Upcoming

There are currently no upcoming events.

    Past

    All Events

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Details:
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping µ: V ⊕ V ∗ → g∗ and we denote µ−1(0) by N. We find criteria for N to have rational singularities and for the categorical quotient N//G to have symplectic singularities. An important special case is V = p g (the direct sum of p copies of g) where p ≥ 2. We show that N has rational singularities and that N//G has symplectic singularities, improving upon results of Aizenbud-Avni and Budur. Let π = π1(Σ) where Σ is a closed Riemann surface of genus p ≥ 2. Let G be semisimple and let Hom(π, G) and X(π, G) be the corresponding representation variety and character variety. The results above imply that Hom(π, G) is a complete intersection with rational singularities and that X(π, G) has symplectic singularities.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: CRITERIA FOR THE ZERO FIBER OF A MOMENT MAP TO HAVE RATIONAL SINGULARITIES, AND ... Read more CRITERIA FOR THE ZERO FIBER OF A MOMENT MAP TO HAVE RATIONAL SINGULARITIES, AND APPLICATIONS. Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    06
    Wednesday
    October
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Finite multiplicities beyond spherical pairs
    Lecturer: Dmitry Gourevitch
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a real reductive algebraic group, and let H be an algebraic subgroup of ... Read more Let G be a real reductive algebraic group, and let H be an algebraic subgroup of G. Itis known that the action of G on the space of functions on G/H is "tame" if this space is spherical. In particular, the multiplicities of the space of Schwartz functions on G/H are finite in this case. I will talk about a recent joint work with A. Aizenbud in which we formulate and analyze a generalization of sphericity that implies finite multiplicities in the Schwartz space of G/H for small enough irreducible smooth representations of G. In more detail, for every G-space X, and every closed G-invariant subset S of the nilpotent cone of the Lie algebra of G, we define when X is S-spherical, by means of a geometric condition involving dimensions of fibers of the moment map. We then show that if X is S-spherical, then every representation with annihilator variety lying in S has (at most) finite multiplicities in the Schwartz space of X. For the case when S is the closure of the Richardson orbit of a parabolic subgroup P of G, we show that the condition is equivalent to P having finitely many orbits on X. We give applications of our results to branching problems. Our main tool in bounding the multiplicity is the theory of holonomic D-modules. After formulating our main results, I will briefly recall the necessary aspects of this theory and sketch our proofs. The talk is based on arXiv:2109.00204.
    Close abstract

    Foundations of Computer Science Colloquium

    Date:
    04
    Monday
    October
    2021
    Lecture / Seminar
    Time: 11:15-12:45
    Title: The Fine-Grained Complexity of Answering Database Queries
    Location: Jacob Ziskind Building
    Lecturer: Nofar Carmeli
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We wish to identify the queries that can be solved with close to optimal time gu ... Read more We wish to identify the queries that can be solved with close to optimal time guarantees over relational databases. Computing all query answers requires at least linear time before the first answer (to read the input and determine the answer's existence), and then we must allow enough time to print all answers (which may be many). Thus, we aspire to achieve linear preprocessing time and constant or logarithmic time per answer. A known dichotomy classifies Conjunctive Queries into those that admit such enumeration and those that do not: the main difficulty of query answering is joining tables, which can be done efficiently if and only if the join query is acyclic. However, the join query usually does not appear in a vacuum
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    20
    Monday
    September
    2021
    Lecture / Seminar
    Time: 11:20
    Title: Criteria for the zero fiber of a moment map to have rational singularities and applications
    Lecturer: Gerald Schwarz
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a complex reductive group with Lie algebra g and let V be a G-module. T ... Read more Let G be a complex reductive group with Lie algebra g and let V be a G-module. There is a natural moment mapping : V  V  ! g and we denote 
    Close abstract

    Geometric Functional Analysis and Probability Seminar

    Date:
    19
    Thursday
    August
    2021
    Lecture / Seminar
    Time: 09:00-10:00
    Title: Lower bounds on the eigenfunctions of random Schroedinger operators in a strip
    Lecturer: Sasha Sodin
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: It is known that the eigenfunctions of a random Schroedinger operator in a stri ... Read more It is known that the eigenfunctions of a random Schroedinger operator in a strip (the direct product of the integer line and a finite set) decay exponentially. In some regimes, the same is true in higher dimensions. It is however not clear whether the eigenfunctions have an exact rate of exponential decay. In the strip, it is natural to expect that the rate should be given by the slowest Lyapunov exponent, however, only the upper bound has been previously established. We shall discuss some recent progress on this problem, and its connection to a question, perhaps interesting in its own right, from the theory of random matrix products. Based on joint work with Ilya Goldsheid.
    Close abstract

    Vision and Robotics Seminar

    Date:
    15
    Sunday
    August
    2021
    Lecture / Seminar
    Time: 09:00-10:30
    Title: Algebraic Characterization of Relational Camera Pose Measurements in Multiple Images
    Lecturer: Yoni Kasten
    Organizer: Faculty of Mathematics and Computer Science
    Details: Structure from Motion (SfM) deals with recovering camera parameters and 3D scene ... Read more Structure from Motion (SfM) deals with recovering camera parameters and 3D scene structure from collections of 2D images. SfM is commonly solved by minimizing the non-covex, bundle adjustment objective, which generally requires sophisticated initialization. In this talk I will present two approaches to SfM: the first approach involves averaging of essential or fundamental matrices (also called bifocal tensors). Since the bifocal tensors are computed independently from image pairs they are generally inconsistent with any set of n cameras.
    Close details
    Abstract: Structure from Motion (SfM) deals with recovering camera parameters and 3D scene ... Read more Structure from Motion (SfM) deals with recovering camera parameters and 3D scene structure from collections of 2D images. SfM is commonly solved by minimizing the non-covex, bundle adjustment objective, which generally requires sophisticated initialization. In this talk I will present two approaches to SfM: the first approach involves “averaging” of essential or fundamental matrices (also called “bifocal tensors”). Since the bifocal tensors are computed independently from image pairs they are generally inconsistent with any set of n cameras. We provide a complete algebraic characterization of the manifold of bifocal tensors for n cameras and present an optimization framework to project measured bifocal tensors onto the manifold. Our second approach is an online approach: given n-1 images, I_1,...,I_{n-1}, whose camera matrices have already been recovered, we seek to recover the camera matrix associated with an image I_n . We present a novel solution to the six-point online algorithm to recover the exterior parameters associated with I_n. Our algorithm uses just six corresponding pairs of 2D points, extracted each from I_n and from any of the preceding n-1 images, allowing the recovery of the full six degrees of freedom of the n'th camera, and unlike common methods, does not require tracking feature points in three or more images. We present experiments that demonstrate the utility of both our approaches. If time permits, I will briefly present additional recent work for solving SfM using deep neural models.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    28
    Wednesday
    July
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: A Stone-von Neumann equivalence for smooth representations and its application to degenerate Whittaker models
    Lecturer: Siddhartha Sahi
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Stone von-Neuman theorem relates the irreducible unitary representations of ... Read more The Stone von-Neuman theorem relates the irreducible unitary representations of the Heisenberg group $H_n$ to non-trivial unitary characters of its center $Z$, and plays a crucial role in the construction of the oscillator representation for the metaplectic group. We give two extensions of this result to non-unitary and non-irreducible representations, thereby obtaining an equivalence of categories between certain representations of $Z$ and those of $H_n$. Our first result is an algebraic equivalence, which can be regarded as a generalization of Kashiwara's lemma from the theory of $D$-modules. Our second result is a smooth equivalence, which involves the fundamental ideas of Ducloux on differentiable representations and smooth imprimitivity systems for Nash groups. We show how to extend the oscillator representation to the smooth setting and give an application to degenerate Whittaker models for representations of reductive groups. This is joint work with Raul Gomez and Dmitry Gourevitch.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    14
    Wednesday
    July
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Certain Poisson Summation Formulae on GL(1) and Langlands Automorphic L-functions
    Lecturer: Dihua Jiang
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: In Tate's famous thesis, harmonic analysis on GL(1) was used to establish the ... Read more In Tate's famous thesis, harmonic analysis on GL(1) was used to establish the global functional equation for Hecke L-functions. In this talk, I will discuss possibilities to define more general Fourier operators and their associated Poisson summation formulae on GL(1), which are expected to be responsible for the global functional equation of general Langlands automorphic L-functions. This is a progress report of my joint work with Zhilin Luo.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    30
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: A number theoretic characterization of (FRS) morphisms: uniform estimates over finite rings of the form Z/p^kZ.
    Lecturer: Yotam Hendel
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let f:X->Y be a morphism between smooth algebraic varieties defined over the int ... Read more Let f:X->Y be a morphism between smooth algebraic varieties defined over the integers. We show its fibers satisfy an extension of the Lang-Weil bounds with respect to finite rings of the form Z/p^kZ uniformly in p, k and in the base point y if and only if f is flat and its fibers have rational singularities, a property abbreviated as (FRS). This characterization of (FRS) morphisms serves as a joint refinement of two results of Aizenbud and Avni
    Close abstract

    Superalgebra Theory and Representations Seminar

    Date:
    30
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 09:00-10:30
    Title: Defect subgroups for algebraic supergroups and super grassmannians
    Location: Jacob Ziskind Building
    Lecturer: Prof. Vera Serganova
    Organizer: Faculty of Mathematics and Computer Science

    Foundations of Computer Science Colloquium

    Date:
    28
    Monday
    June
    2021
    Lecture / Seminar
    Time: 09:15-10:30
    Title: Distributed Subgraph Finding: Progress and Challenges
    Location: Jacob Ziskind Building
    Lecturer: Keren Censor-Hillel
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: This talk surveys the exciting recent progress made in understanding the complex ... Read more This talk surveys the exciting recent progress made in understanding the complexity of distributed subgraph finding problems. I will discuss techniques and highlight some intriguing open problems.
    Close abstract

    Foundations of Computer Science Colloquium

    Date:
    21
    Monday
    June
    2021
    Lecture / Seminar
    Time: 09:15-10:30
    Title: LDPC Codes Achieve List Decoding Capacity
    Location: Jacob Ziskind Building
    Lecturer: Noga Ron-Zewi
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We show that random Low-Density Parity Check (LDPC) codes achieve list decoding ... Read more We show that random Low-Density Parity Check (LDPC) codes achieve list decoding capacity with high probability. These are the first graph-based codes shown to have this property. This result follows from two other more general results that may be of independent interest: 1. A simple characterization of the threshold for when ‘local’ combinatorial properties are likely to be satisfied by a random subspace over a finite field. 2. Any ‘local’ property that is satisfied with high probability by a random linear code is also satisfied with high probability by a random LDPC code. Based on joint work with Jonathan Mosheiff, Nicolas Resch, Shashwat Silas, and Mary Wootters.
    Close abstract

    Vision and Robotics Seminar

    Date:
    17
    Thursday
    June
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Combining bottom-up and top-down computations for image interpretation
    Lecturer: Hila Levi
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Visual scene understanding has traditionally focused on identifying objects in i ... Read more Visual scene understanding has traditionally focused on identifying objects in images – learning to predict their presence and spatial extent. However, understanding a visual scene often goes beyond recognizing individual objects. In my thesis (guided by Prof. Ullman), I mainly focused on developing `task-dependent network’, that uses processing instructions to guide the functionality of a shared network via an additional task input. In addition, I also studied strategies for incorporating relational information into recognition pipelines to efficiently extract structures of interest from the scene. In the scope of high level scene understanding, which might be dominated by recognizing a rather small number of objects and relations, the above task-dependent scheme naturally allows goal-directed scene interpretation by either a single step or by a sequential execution with a series of different TD instructions. It simplifies the use of referring relations, grounds requested visual concepts back into the image-plane and improves combinatorial generalization, essential for AI systems, by using structured representations and computations. In the scope of ‘multi-task learning’ the above scheme offers an alternative to the popular ‘multi-branched architecture’, which simultaneously execute all tasks using task-specific branches on top of a shared backbone, challenges capacity limitations, increases task selectivity, allows scalability and further tasks extensions. Results will be shown in various applications: object detection, visual grounding, properties classification, human-object interactions and general scene interpretation. Works included: 1. H. Levi and S. Ullman. Efficient coarse-to-fine non-local module for the detection of small objects. BMVC, 2019. https://arxiv.org/abs/1811.12152 2. H. Levi and S. Ullman. Multi-task learning by a top-down control network. ICIP 2021. https://arxiv.org/abs/2002.03335 3. S. Ullman et al. Image interpretation by iterative bottom-up top-down processing. http://arxiv.org/abs/2105.05592 4. A. Arbelle et al. Detector-Free Weakly Supervised Grounding by Separation. https://arxiv.org/abs/2104.09829. Submitted to ICCV. 5. Ongoing work – Human Object Interactions
    Close abstract

    Vision and Robotics Seminar

    Date:
    17
    Thursday
    June
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Combining bottom-up and top-down computations for image interpretation
    Lecturer: Hila Levi
    Organizer: Faculty of Mathematics and Computer Science
    Details: Visual scene understanding has traditionally focused on identifying objects in images
    Abstract: Visual scene understanding has traditionally focused on identifying objects in i ... Read more Visual scene understanding has traditionally focused on identifying objects in images – learning to predict their presence and spatial extent. However, understanding a visual scene often goes beyond recognizing individual objects. In my thesis (guided by Prof. Ullman), I mainly focused on developing task-dependent network, that uses processing instructions to guide the functionality of a shared network via an additional task input. In addition, I also studied strategies for incorporating relational information into recognition pipelines to efficiently extract structures of interest from the scene. In the scope of high level scene understanding, which might be dominated by recognizing a rather small number of objects and relations, the above task-dependent scheme naturally allows goal-directed scene interpretation by either a single step or by a sequential execution with a series of different TD instructions. It simplifies the use of referring relations, grounds requested visual concepts back into the image-plane and improves combinatorial generalization, essential for AI systems, by using structured representations and computations. In the scope of multi-task learning the above scheme offers an alternative to the popular multi-branched architecture, which simultaneously execute all tasks using task-specific branches on top of a shared backbone, challenges capacity limitations, increases task selectivity, allows scalability and further tasks extensions. Results will be shown in various applications: object detection, visual grounding, properties classification, human-object interactions and general scene interpretation. Works included: 1. H. Levi and S. Ullman. Efficient coarse-to-fine non-local module for the detection of small objects. BMVC, 2019. https://arxiv.org/abs/1811.12152 2. H. Levi and S. Ullman. Multi-task learning by a top-down control network. ICIP 2021. https://arxiv.org/abs/2002.03335 3. S. Ullman et al. Image interpretation by iterative bottom-up top-down processing. http://arxiv.org/abs/2105.05592 4. A. Arbelle et al. Detector-Free Weakly Supervised Grounding by Separation. https://arxiv.org/abs/2104.09829. Submitted to ICCV. 5. Ongoing work – Human Object Interactions
    Close abstract

    Vision and Robotics Seminar

    Date:
    17
    Thursday
    June
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Combining bottom-up and top-down computations for image interpretation
    Lecturer: Hila Levi
    Organizer: Faculty of Mathematics and Computer Science
    Details: Visual scene understanding has traditionally focused on identifying objects in i ... Read more Visual scene understanding has traditionally focused on identifying objects in images – learning to predict their presence and spatial extent. However, understanding a visual scene often goes beyond recognizing individual objects. In my thesis (guided by Prof. Ullman), I mainly focused on developing `task-dependent network’, that uses processing instructions to guide the functionality of a shared network via an additional task input. In addition, I also studied strategies for incorporating relational information into recognition pipelines to efficiently extract structures of interest from the scene. In the scope of high level scene understanding, which might be dominated by recognizing a rather small number of objects and relations, the above task-dependent scheme naturally allows goal-directed scene interpretation by either a single step or by a sequential execution with a series of different TD instructions. It simplifies the use of referring relations, grounds requested visual concepts back into the image-plane and improves combinatorial generalization, essential for AI systems, by using structured representations and computations. In the scope of ‘multi-task learning’ the above scheme offers an alternative to the popular ‘multi-branched architecture’, which simultaneously execute all tasks using task-specific branches on top of a shared backbone, challenges capacity limitations, increases task selectivity, allows scalability and further tasks extensions. Results will be shown in various applications: object detection, visual grounding, properties classification, human-object interactions and general scene interpretation. Works included: 1. H. Levi and S. Ullman. Efficient coarse-to-fine non-local module for the detection of small objects. BMVC, 2019. https://arxiv.org/abs/1811.12152 2. H. Levi and S. Ullman. Multi-task learning by a top-down control network. ICIP 2021. https://arxiv.org/abs/2002.03335 3. S. Ullman et al. Image interpretation by iterative bottom-up top-down processing. http://arxiv.org/abs/2105.05592 4. A. Arbelle et al. Detector-Free Weakly Supervised Grounding by Separation. https://arxiv.org/abs/2104.09829. Submitted to ICCV. 5. Ongoing work – Human Object Interactions
    Close details
    Abstract: ... Read more
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    16
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 14:30-16:00
    Title: Multiplicity one theorems over positive characteristic
    Location: Jacob Ziskind Building
    Lecturer: Dor Mezer
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We will discuss the recent proof of the strong Gelfand property over local field ... Read more We will discuss the recent proof of the strong Gelfand property over local fields of positive odd characteristic for the Gan-Gross-Prasad pairs. These pairs of groups include (GL(n),GL(n 1)),(O (n),O(n 1)), (U(n),U(n 1)), (SO(n),SO(n 1)), as well as Fourier-Jacobi pairs such as the symplectic group inside its semi-direct product with the corresponding Heisenberg group. The strong Gelfand property for these pairs has important consequences in the theory of automorphic representations, showing the uniqueness of Bessel models, of Rankin-Selberg models, and of Fourier-Jacobi models. These results were proven over characteristic 0 local fields (both archimedean and p-adic) 10-14 years ago, and extensively used since then. As in the characteristic zero case, we will use the method of Gelfand and Kazhdan, which reduces the theorems to statements on invariant distributions.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    16
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 14:30-15:25
    Title: Multiplicity one theorems over positive characteristic
    Organizer: Faculty of Mathematics and Computer Science

    Foundations of Computer Science Colloquium

    Date:
    14
    Monday
    June
    2021
    Lecture / Seminar
    Time: 09:15-10:30
    Title: What Is The Sample Complexity of Differentially Private Learning?
    Location: Jacob Ziskind Building
    Lecturer: Shay Moran
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The increase in machine learning applications which involve private and personal ... Read more The increase in machine learning applications which involve private and personal data highlights the need for algorithms that handle the data *responsibly*. While this need has been successfully addressed by the field of differentially private machine learning, the cost of privacy remains poorly understood: How much data is needed for differentially private learning? How much more data does private learning require compared to learning without privacy constraints? We will survey some of the recent progress towards answering these questions in the distribution-free PAC model, including the Littlestone-dimension-based *qualitative* characterization and the relationship with online learning. If time allows, we will also discuss this question in more general (distribution- and data-dependent) learning models. Based on joint works with Noga Alon, Mark Bun, Roi Livni, and Maryanthe Malliaris
    Close abstract

    Vision and Robotics Seminar

    Date:
    10
    Thursday
    June
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Drop the GAN: In Defense of Patches Nearest Neighbors as Single Image Generative Models
    Lecturer: Niv Granot
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Single image generative models perform synthesis and manipulation tasks by captu ... Read more Single image generative models perform synthesis and manipulation tasks by capturing the distribution of patches within a single image. The classical (pre Deep Learning) prevailing approaches for these tasks are based on an optimization process that maximizes patch similarity between the input and generated output. Recently, however, Single Image GANs were introduced both as a superior solution for such manipulation tasks, but also for remarkable novel generative tasks. Despite their impressiveness, single image GANs require long training time (usually hours) for each image and each task. They often suffer from artifacts and are prone to optimization issues such as mode collapse. We show that all of these tasks can be performed without any training, within several seconds, in a unified, surprisingly simple framework. The "good-old" patch-based methods are revisited and casted into a novel optimization-free framework. This allows generating random novel images better and much faster than GANs. We further demonstrate a wide range of applications, such as image editing and reshuffling, retargeting to different sizes, structural analogies, image collage and a newly introduced task of conditional inpainting. Not only is our method faster (×103-×104 than a GAN), it produces superior results (confirmed by quantitative and qualitative evaluation), less artifacts and more realistic global structure than any of the previous approaches (whether GAN-based or classical patch-based).
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    09
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Representations of a reductive $p$-adic group $G$ over a field $C$.
    Lecturer: Marie-France Vignéras
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: When $C$ is algebraically closed of characteristic different from $p$, for man ... Read more When $C$ is algebraically closed of characteristic different from $p$, for many groups $G$, a list of pairs $(J,lambda)$, where $lambda$ is a smooth $C$-representation of a compact modulo centre subgroup $J$ of $G$, has been produced such that any irreducible cuspidal $C$-representation of $G$ has the form $ ind_J^Glambda$ , for a pair $(J, lambda)$ unique up to conjugation. With Guy Henniart, we produced similar lists when $C$ is no longer assumed algebraically closed. Our other main result concerns supercuspidality. The notion of supercuspidality makes sense for the irreducible cuspidal $C$-representations of $G$, and also for the representations $lambda$. In most cases we proved that $ind^G_Jlambda$ is supercuspidal if and only if $lambda$ is supercuspidal.
    Close abstract

    Vision and Robotics Seminar

    Date:
    03
    Thursday
    June
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: StyleSpace Analysis: Disentangled Controls for StyleGAN Image Generation
    Lecturer: Zongze Wu
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We explore and analyze the latent style space of StyleGAN2, a state-of-the-art a ... Read more We explore and analyze the latent style space of StyleGAN2, a state-of-the-art architecture for image generation, using models pretrained on several different datasets. We first show that StyleSpace, the space of channel-wise style parameters, is significantly more disentangled than the other intermediate latent spaces explored by previous works. Next, we describe a method for discovering a large collection of style channels, each of which is shown to control a distinct visual attribute in a highly localized and disentangled manner. Third, we propose a simple method for identifying style channels that control a specific attribute, using a pretrained classifier or a small number of example images. Manipulation of visual attributes via these StyleSpace controls is shown to be better disentangled than via those proposed in previous works. To show this, we make use of a newly proposed Attribute Dependency metric. Finally, we demonstrate the applicability of StyleSpace controls to the manipulation of real images. Our findings pave the way to semantically meaningful and well-disentangled image manipulations via simple and intuitive interfaces. This is a joint work with Dani Lischinski and Eli Shechtman. paper(CVPR 2021 oral): https://arxiv.org/abs/2011.12799 code: https://github.com/betterze/StyleSpace video: https://www.youtube.com/watch?v=U7qRotRGr1w
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    02
    Wednesday
    June
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: On the local coefficients matrix for covering groups and the restriction problem
    Location: Jacob Ziskind Building
    Lecturer: Dani Szpruch
    Organizer: Faculty of Mathematics and Computer Science
    Details: The local coefficients matrix is an analog of Shahidi local coefficients defined ... Read more The local coefficients matrix is an analog of Shahidi local coefficients defined for covering groups in aproject with Fan Gao and Freydoon Shasetting where uniqueness of Whittaker model fails. After introducing this object we shall discuss in some detail our results for coverings of SL(2) and GL(2) and point out a difference between the results for these two groups. Then we shall show how this difference is explained by the nature of the restriction of genuine representations as well as by the subtle relations among certain arithmetic factors. Finally, as time permits, we shall discuss some implications of the restriction mentioned above. This is a joint hidi.
    Close details
    Abstract: The local coefficients matrixis an analog of Shahidi local coefficients defined ... Read more The local coefficients matrixis an analog of Shahidi local coefficients defined for covering groups in asetting where uniqueness of Whittaker model fails. After introducing thisobject we shall discuss in some detail our results for coverings of SL(2) andGL(2) and point out a difference between the results for these two groups. Thenwe shall show how this difference is explained by the nature of the restrictionof genuine representations as well as by the subtle relations among certainarithmetic factors. Finally, as time permits, we shall discuss someimplications of the restriction mentioned above. This is a joint project withFan Gao and Freydoon Shahidi.
    Close abstract

    Foundations of Computer Science Colloquium

    Date:
    24
    Monday
    May
    2021
    Lecture / Seminar
    Time: 09:15-10:15
    Title: Non-adaptive vs Adaptive Queries in the Dense Graph Testing Model by Oded Goldreich and Avi Wigderson
    Location: Jacob Ziskind Building
    Lecturer: Oded Goldreich
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We study the relation between the query complexity of adaptive and non-adaptive ... Read more We study the relation between the query complexity of adaptive and non-adaptive testers in the dense graph model. It has been known for a couple of decades that the query complexity of non-adaptive testers is at most quadratic in the query complexity of adaptive testers. We show that this general result is essentially tight
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    19
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Anabelian construction of phi,Gamma modules
    Lecturer: Nadav Gropper
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Anabelian geometry asks how much can we say about a variety from its fundamental ... Read more Anabelian geometry asks how much can we say about a variety from its fundamental group. In 1997 Shinichi Mochizuki, using p adic Hodge theory, proved a fundamental anabelian result for the case of p-adic fields. In my talk I will discuss representation theoretical data which can be reconstructed from an absolute Galois group and also types of representations that cannot be constructed solely from it. I will also sketch how these types of ideas can potentially give many new results about p-adic Galois representation
    Close abstract

    Vision and Robotics Seminar

    Date:
    13
    Thursday
    May
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Deep Internal Learning
    Lecturer: Assaf Shocher
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Deep Learning has always been divided into two phases: Training and Inference. ... Read more Deep Learning has always been divided into two phases: Training and Inference. The common practice for Deep Learning is training big networks on huge datasets. While very successful, such networks are only applicable to the type of data they were trained for and require huge amounts of annotated data, which in many cases are not available. In my thesis (guided by Prof. Irani), I invented ``Deep Internal Learning''. Instead of learning to generally solve a task for all inputs, we perform ``ad hoc'' learning for specific input. We train an image-specific network, we do it at test-time and on the test-input only, in an unsupervised manner (no label or ground-truth). In this regime, training is actually a part of the inference, no additional data or prior training is taking place. I will demonstrate how we applied this framework for various challenges: Super-Resolution, Segmentation, Dehazing, Transparency-Separation, Watermark removal. I will also show how this approach can be incorporated to Generative Adversarial Networks by training a GAN on a single image. If time permits I will also cover some partially related works. Links to papers: http://www.wisdom.weizmann.ac.il/~vision/zssr http://www.wisdom.weizmann.ac.il/~vision/DoubleDIP http://www.wisdom.weizmann.ac.il/~vision/ingan http://www.wisdom.weizmann.ac.il/~vision/kernelgan https://semantic-pyramid.github.io/ https://arxiv.org/abs/2006.11120 https://arxiv.org/abs/2103.15545
    Close abstract

    Vision and Robotics Seminar

    Date:
    13
    Thursday
    May
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Deep Internal Learning
    Lecturer: Assaf Shocher
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Deep Learning has always been divided into two phases: Training and Inference. ... Read more Deep Learning has always been divided into two phases: Training and Inference. The common practice for Deep Learning is training big networks on huge datasets. While very successful, such networks are only applicable to the type of data they were trained for and require huge amounts of annotated data, which in many cases are not available. In my thesis (guided by Prof. Irani), I invented ``Deep Internal Learning''. Instead of learning to generally solve a task for all inputs, we perform ``ad hoc'' learning for specific input. We train an image-specific network, we do it at test-time and on the test-input only, in an unsupervised manner (no label or ground-truth). In this regime, training is actually a part of the inference, no additional data or prior training is taking place. I will demonstrate how we applied this framework for various challenges: Super-Resolution, Segmentation, Dehazing, Transparency-Separation, Watermark removal. I will also show how this approach can be incorporated to Generative Adversarial Networks by training a GAN on a single image. If time permits I will also cover some partially related works. Links to papers: http://www.wisdom.weizmann.ac.il/~vision/zssr http://www.wisdom.weizmann.ac.il/~vision/DoubleDIP http://www.wisdom.weizmann.ac.il/~vision/ingan http://www.wisdom.weizmann.ac.il/~vision/kernelgan https://semantic-pyramid.github.io/ https://arxiv.org/abs/2006.11120 https://arxiv.org/abs/2103.15545
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: The Burger-Sarnak Method and Operations on the Unitary Duals of Classical Groups
    Lecturer: Andrew Hendrickson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Burger-Sarnak method shows that the restriction of an automorphic representa ... Read more The Burger-Sarnak method shows that the restriction of an automorphic representation of a reductive group to a reductive subgroup has automorphic support. Clozel has conjectured a qualitative refinement of this result, which was first verified and quantified in the GLn case by Venkatesh. In this talk I will describe my thesis which extended this result to classical groups.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: The Burger-Sarnak Method and Operations on the Unitary Duals of Classical Groups
    Lecturer: Andrew Hendrickson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Burger-Sarnak method shows that the restriction of an automorphic representa ... Read more The Burger-Sarnak method shows that the restriction of an automorphic representation of a reductive group to a reductive subgroup has automorphic support. Clozel has conjectured a qualitative refinement of this result, which was first verified and quantified in the GLn case by Venkatesh. In this talk I will describe my thesis which extended this result to classical groups.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: The Burger-Sarnak Method and Operations on the Unitary Duals of Classical Groups
    Lecturer: Andrew Hendrickson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Burger-Sarnak method shows that the restriction of an automorphic representa ... Read more The Burger-Sarnak method shows that the restriction of an automorphic representation of a reductive group to a reductive subgroup has automorphic support. Clozel has conjectured a qualitative refinement of this result, which was first verified and quantified in the GLn case by Venkatesh. In this talk I will describe my thesis which extended this result to classical groups.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: The Burger-Sarnak Method and Operations on the Unitary Duals of Classical Groups
    Lecturer: Andrew Hendrickson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Burger-Sarnak method shows that the restriction of an automorphic representa ... Read more The Burger-Sarnak method shows that the restriction of an automorphic representation of a reductive group to a reductive subgroup has automorphic support. Clozel has conjectured a qualitative refinement of this result, which was first verified and quantified in the GLn case by Venkatesh. In this talk I will describe my thesis which extended this result to classical groups.
    Close abstract

    Vision and Robotics Seminar

    Date:
    06
    Thursday
    May
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: What, How and When can we Learn Adversarially Robustly?
    Lecturer: Nathan Srebro
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: In this talk we will discuss the problem of learning an adversarially robust pre ... Read more In this talk we will discuss the problem of learning an adversarially robust predictor from clean training data. That is, learning a predictor that performs well not only on future test instances, but also when these instances are corrupted adversarially. There has been much empirical interest in this question, and in this talk we will take a theoretical perspective and see how it leads to practically relevant insights, including: the need to depart from an empirical (robust) risk minimization approach, and thinking of what kind of accesses and reductions can allow learning. Joint work with Omar Montasser and Steve Hanneke.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    05
    Wednesday
    May
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Tempered dual and crossed products for real and p-adic reductive groups
    Lecturer: Anne-Marie Aubert
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Let G be a real or a p-adic connected reductive group. We will consider the conn ... Read more Let G be a real or a p-adic connected reductive group. We will consider the connected components of the tempered dual of G. They are labelled by the G-conjugacy classes of pairs formed by a Levi subgroup M of G and the orbit of a discrete series representation of M under the group of unitary unramified characters of M. For real groups, Wassermann proved in 1987, by noncommutative-geometric methods, that each connected component has a simple geometric structure which encodes the reducibility of the corresponding parabolically induced representations. We will explain how one can recover his result. For p-adic groups, each connected component comes with a compact torus equipped with a finite group action, and an analogous result, that we will describe, holds true under a certain geometric assumption on the structure of stabilizers for that action. In the case when G is a quasi-split symplectic, special orthogonal or unitary group, it is possible to explicitly determine the connected components for which the geometric assumption is satisfied. It is a joint work with Alexandre Afgoustidis.
    Close abstract

    Vision and Robotics Seminar

    Date:
    29
    Thursday
    April
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: Robustifying neural networks
    Lecturer: Raja Giryes
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: In this talk I will survey several techniques to make neural networks more robus ... Read more In this talk I will survey several techniques to make neural networks more robust. While neural networks achieve groundbreaking results in many applications, they depend strongly on the availability of good training data and the assumption that the data in the test time will resemble the one at train time. In this talk, we will survey different techniques that we developed for improving the network robustness and/or adapting it to the data at hand.
    Close abstract

    Vision and Robotics Seminar

    Date:
    22
    Thursday
    April
    2021
    Lecture / Seminar
    Time: 10:15-11:30
    Title: A New Theory of Adversarial Examples in Machine Learning
    Lecturer: Adi Shamir
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The extreme fragility of deep neural networks when presented with tiny perturbat ... Read more The extreme fragility of deep neural networks when presented with tiny perturbations in their inputs was independently discovered by several research groups in 2013. Due to their mysterious properties and major security implications, these adversarial examples had been studied extensively over the last eight years, but in spite of enormous effort they remained a baffling phenomenon with no clear explanation. In particular, it was not clear why a tiny distance away from almost any cat image there are images which are recognized with a very high level of confidence as cars, planes, frogs, horses, or any other desired class, why the adversarial modification which turns a cat into a car does not look like a car at all, and why a network which was adversarially trained with randomly permuted labels (so that it never saw any image which looks like a cat being called a cat) still recognizes most cat images as cats. The goal of this talk is to introduce a new theory of adversarial examples, which we call the Dimpled Manifold Model. It can easily explain in a simple and intuitive way why they exist and why they have all the bizarre properties mentioned above. In addition, it sheds new light on broader issues in machine learning such as what happens to deep neural networks during regular and during adversarial training. Experimental support for this theory, obtained jointly with Oriel Ben Shmuel and Odelia Melamed, will be presented and discussed in the last part of the talk.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    07
    Wednesday
    April
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: A reduction principle for Fourier coefficients of automorphic forms
    Lecturer: Siddhartha Sahi
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We consider a general class of Fourier coefficients for an automorphic form on a ... Read more We consider a general class of Fourier coefficients for an automorphic form on a finite cover of a reductive adelic group G(A_K), associated to the data of a `Whittaker pair'. We describe a quasi-order on Fourier coefficients, and an algorithm that gives an explicit formula for any coefficient in terms of integrals and sums involving higher coefficients. The maximal elements for the quasi-order are `Levi-distinguished' Fourier coefficients, which correspond to taking the constant term along the unipotent radical of a parabolic subgroup, and then further taking a Fourier coefficient with respect to a K-distinguished nilpotent orbit in the Levi quotient. Thus one can express any Fourier coefficient, including the form itself, in terms of higher Levi-distinguished coefficients. In follow-up papers we use this result to determine explicit Fourier expansions of minimal and next-to-minimal automorphic forms on split simply-laced reductive groups, and to obtain Euler product decompositions of their top Fourier coefficients. This is joint work with Dmitry Gourevitch, Henrik P. A. Gustafsson, Axel Kleinschmidt, and Daniel Persson https://arxiv.org/abs/1811.05966
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    17
    Wednesday
    March
    2021
    Lecture / Seminar
    Time: 16:30-17:30
    Title: The orbit method, microlocal analysis and applications to L-functions
    Lecturer: Paul Nelson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: I will describe how the orbit method can be developed in a quantitative form, al ... Read more I will describe how the orbit method can be developed in a quantitative form, along the lines of microlocal analysis, and applied to local problems in representation theory and global problems concerning automorphic forms. The local applications include asymptotic expansions of relative characters. The global applications include moment estimates and subconvex bounds for L-functions. These results are the subject of two papers, the first joint with Akshay Venkatesh: https://arxiv.org/abs/1805.07750 https://arxiv.org/abs/2012.0218
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    10
    Wednesday
    March
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Weyl group representations and Harish-Chandra cells
    Lecturer: David Vogan
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Suppose g is a semisimple Lie algebra with Weyl group W. Write L(w) for the irre ... Read more Suppose g is a semisimple Lie algebra with Weyl group W. Write L(w) for the irreducible highest weight module of highest weight -w.rho - rho. Write J (for "Joseph") for the set of primitive ideals in a semisimple enveloping algebra contained in the augmentation ideal. In a 1978 paper "W-module structure in the primitive spectrum..." Joseph attached to each primitive ideal I in J a subset Lcell(I) = {w in W | Ann(L(w)) = I}. He showed also how to make Lcell(I) into a basis for a representation sigma(I) of W, in such a way that sum_{I in J} sigma(I) = regular representation of W. These representations sigma(I) are now called "left cell representations," terminology that is apparently due to Joseph (see his 1981 paper "Goldie rank in the enveloping algebra...III," page 310). Joseph proved in a 1980 paper that each left cell representation consists of exactly one copy of Joseph's "Goldie rank representation" for the primitive ideal I, and some additional representations that are not Goldie rank representations. For the past forty years, understanding of these left cell representations of W has been at the heart of a great deal of work on representations of reductive groups. Lusztig in his 1984 book gave a description of all left cells in terms of the geometry of nilpotent orbits. Part of Lusztig's description uses Springer's parametrization of W representations by irreducible representations of the equivariant fundamental group A(O) for a nilpotent orbit O. I will discuss the "opposite" part of Lusztig's description, involving conjugacy classes in A(O).
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    03
    Wednesday
    March
    2021
    Lecture / Seminar
    Time: 14:30-15:30
    Title: Period Relations of Standard $L$-Functions of Symplectic Type
    Lecturer: Tian Fangyang
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: A classical result of Euler says that the value of the Riemann-Zeta function at ... Read more A classical result of Euler says that the value of the Riemann-Zeta function at a positive even integer $2k$ is a rational multiple of $pi^{2k}$. In the 1970s, a successive pioneering work of G. Shimura revealed the relation of different critical values of $L$-function that are attached to modular forms of $mathrm{GL}_2$. This type of result, conjectured by D. Blasius for general linear groups, is called period relation of a certain automorphic $L$-function, which is closely related to a celebrated conjecture of P. Deligne. In this talk, I will discuss my work joint with Dihua Jiang and Binyong Sun on the period relation for the twisted standard L-function $L(s, Piotimeschi)$, where $Pi$ is an irreducible cuspidal automorphic representation of $GL_{2n}(mathbb{A})$ which is regular algebraic and of symplectic type. Along this talk, I will also discuss the key ingredient of this project - the existence of uniform cohomological test vector, which provides the most precise information on the archimedean local integrals.
    Close abstract

    Foundations of Computer Science Colloquium

    Date:
    01
    Monday
    March
    2021
    Lecture / Seminar
    Time: 08:30
    Title: Rounding Dynamic Matchings Against an Adaptive Adversary
    Lecturer: David Wajc
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: Dynamic algorithms address dynamically-evolving datasets, and show how to mainta ... Read more Dynamic algorithms address dynamically-evolving datasets, and show how to maintain solutions to problems of interest subject to small updates to the input much faster than re-computing these solutions from scratch after each update. For example, the dynamic matching problem asks us to maintain an approximate maximum matching under edge updates (additions deletions). For this well-studied problem we know numerous dynamic algorithms, the fastest of which are randomized. Unfortunately, until recently, all these fast randomized algorithms assumed that the updates to the input are generated in advance, rather than adaptively, based on the algorithm's random coin tosses. This assumption, referred to as the oblivious adversary assumption in the literature, rules out the black-box use of these algorithms for speeding up other (dynamic or static) algorithms. In this talk, I will present my recent framework for obtaining (essentially) the same randomized bounds previously known to be obtainable against an oblivious adversary, but this time against an adaptive adversary.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    24
    Wednesday
    February
    2021
    Lecture / Seminar
    Time: 16:30
    Title: Recent progress on the Gan-Gross-Prasad conjecture for unitary groups
    Lecturer: Pierre-Henri Chaudouard
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The talk is based on a joint work with Raphaël Beuzart-Plessis and Michal Zydo ... Read more The talk is based on a joint work with Raphaël Beuzart-Plessis and Michal Zydor and on an ongoing joint work with Raphaël Beuzart-Plessis. The Gan-Gross-Prasad (GGP) conjecture relates the non-vanishing of some periods of cuspidal automorphic forms to that of central values of some related L-functions. In the case of unitary groups U(n)xU(n 1), a lot of progress have been made by a deep study of the Jacquet-Rallis trace formula. We will describe some recent results that can be obtained by a further analysis of the trace formula. Besides the "endoscopic cases" of the GGP conjecture, we get an extension to the periods of some Eisenstein series. We will also discuss applications we can get to the so-called Bessel periods of unitary groups.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    21
    Sunday
    February
    2021
    -
    17
    Wednesday
    March
    2021
    Lecture / Seminar
    Time: 09:00 - 16:30
    Title: The orbit method,microlocal analysis and applications to L-functions
    Lecturer: Paul Nelson
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: I will describe how the orbit method can be developed in a quantitative form, al ... Read more I will describe how the orbit method can be developed in a quantitative form, along the lines of microlocal analysis, and applied to local problems in representation theory and global problems concerning automorphic forms. The local applications include asymptotic expansions of relative characters. The global applications include moment estimates and subconvex bounds for L-functions. These results are the subject of two papers, the first joint with Akshay Venkatesh:
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    17
    Wednesday
    February
    2021
    Lecture / Seminar
    Time: 16:30-17:30
    Title: The generalized doubling method, multiplicity one and applications.
    Lecturer: Eyal Kaplan
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The doubling method of Piatetski-Shapiro and Rallis pioneered the study of integ ... Read more The doubling method of Piatetski-Shapiro and Rallis pioneered the study of integral representations of automorphic L-functions, for cuspidal representations (generic or otherwise) of classical groups. Originating in the calculation of the Petersson inner product of the theta lift, this method has been successfully applied within the theory of the theta correspondence, and had numerous additional applications to the theory of L-functions and to arithmetic problems. Recently, the doubling method has been generalized in several aspects with interesting applications to global functoriality, automorphic descent and the study of representations of covering groups. In this talk I will survey the different components of the generalized doubling method, focusing on one of the fundamental results: Local multiplicity one, obtained recently in a joint work with Dima and Rami. I will also describe a new GL(c) x GL(k) doubling type integral which interpolates between the integrals of Godement and Jacquet and the integrals of Jacquet, Piatetski-Shapiro and Shalika. This integral was used in order to obtain certain poles within the doubling construction. Parts of the talk are also based on a collaboration with Cai, Friedberg and Ginzburg.
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    03
    Wednesday
    February
    2021
    Lecture / Seminar
    Time: 16:30-17:30
    Title: What Atlas can do
    Lecturer: Jeffrey Adams
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The Atlas of Lie groups and representations is a project to use computational me ... Read more The Atlas of Lie groups and representations is a project to use computational methods in representation theory, and in particular to compute the unitary dual. Our algorithms are implemented in the atlas software (www.liegroups.org). Besides computing unitary representations, the software is useful in performing a wide variety of calculations in Lie theory. I will give some examples of what the software can do. Examples include: geometry of the action of symmetric subgroups on the flag variety
    Close abstract

    Vision and Robotics Seminar

    Date:
    28
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Learning to Sample
    Lecturer: Shai Avidan
    Organizer: Faculty of Mathematics and Computer Science
    Details: There is a growing number of tasks that work directly on point clouds. As the si ... Read more There is a growing number of tasks that work directly on point clouds. As the size of the point cloud grows, so do the computational demands of these tasks. A possible solution is to sample the point cloud first. Classic sampling approaches, such as farthest point sampling (FPS), do not consider the downstream task. A recent work showed that learning a task-specific sampling can improve results significantly. However, the proposed technique did not deal with the non-differentiability of the sampling operation and offered a workaround instead. We introduce a novel differentiable relaxation for point cloud sampling that approximates sampled points as a mixture of points in the primary input cloud. Our approximation scheme leads to consistently good results on classification and geometry reconstruction applications. We also show that the proposed sampling method can be used as a front to a point cloud registration network. This is a challenging task since sampling must be consistent across two different point clouds for a shared downstream task. In all cases, our approach outperforms existing non-learned and learned sampling alternatives. Based on the work of: Itai Lang, Oren Dovrat, and Asaf Manor
    Close details
    Abstract: ... Read more
    Close abstract

    Superalgebra Theory and Representations Seminar

    Date:
    27
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Bigrassmannian permutations and Verma modules
    Lecturer: Volodymyr Mazorchuk
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk I will describe how bigrassmannian permutations control the socle o ... Read more In this talk I will describe how bigrassmannian permutations control the socle of the cokernel of embeddings of Verma modules for sl_n. An application of this is a description of the socle of the cokernel of homomorphisms between Verma modules for the periplective Lie superalgebra. This is based on two joint works: one with Hankyung Ko and Rafael Mrden and another one with Chih-Whi Chen.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    27
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 14:00-15:30
    Title: Deep Learning on Structured and Geometric Data
    Lecturer: Haggai Maron
    Organizer: Faculty of Mathematics and Computer Science
    Details: Deep Learning of structured and geometric data, such as sets, graphs, and surfac ... Read more Deep Learning of structured and geometric data, such as sets, graphs, and surfaces, is a prominent research direction that has received considerable attention in the last few years. Given a learning task that involves structured data, the main challenge is identifying suitable neural network architectures and understanding their theoretical and practical tradeoffs. This talk will focus on a popular learning setup where the learning task is invariant to a group of transformations of the input data. This setup is relevant to many popular learning tasks and data types. In the first part of the talk, I will present a general framework for designing neural network architectures based on layers that respect these transformations. In particular, I will show that these layers can be implemented using parameter-sharing schemes induced by the group. In the second part of the talk, I will demonstrate the framework’s applicability by presenting novel neural network architectures for two widely used data types: graphs and sets. I will also show that these architectures have desirable theoretical properties and that they perform well in practice.
    Close details
    Abstract: ... Read more
    Close abstract

    Faculty Seminar

    Date:
    26
    Tuesday
    January
    2021
    Lecture / Seminar
    Time: 16:00-17:30
    Title: Towards Reliable Data-Driven Computations
    Lecturer: Yuval Moskovitch
    Organizer: Faculty of Mathematics and Computer Science
    Details: Data-driven methods are increasingly being used in domains such as fraud and ris ... Read more Data-driven methods are increasingly being used in domains such as fraud and risk detection, where data-driven algorithmic decision making may affect human life. The growing impact of data and data-driven systems on society makes it important that people be able to trust analytical results obtained from data-driven computations. This can be done in two complementary ways: by providing result explanations so that the user understands the computation and the basis for the observed results
    Close details

    Machine Learning and Statistics Seminar

    Date:
    21
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 18:00-19:00
    Title: Learning on Pointclouds for 3D Scene Understanding
    Lecturer: Or Litany
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk i'll be covering several works in the topic of 3D deep learning on ... Read more In this talk i'll be covering several works in the topic of 3D deep learning on pointclouds for scene understanding tasks. First, I'll describe VoteNet (ICCV 2019, best paper nomination): a method for object detection from 3D pointclouds input, inspired by the classical generalized Hough voting technique. I'll then explain how we integrated image information into the voting scheme to further boost 3D detection (ImVoteNet, CVPR 2020). In the second part of my talk I'll describe recent studies focusing on reducing supervision for 3D scene understanding tasks, including PointContrast -- a self-supervised representation learning framework for 3D pointclods (ECCV 2020). Our findings in PointContrast are extremely encouraging: using a unified triplet of architecture, source dataset, and contrastive loss for pre-training, we achieve improvement over recent best results in segmentation and detection across 6 different benchmarks for indoor and outdoor, real and synthetic datasets -- demonstrating that the learned representation can generalize across domains.
    Close details

    Vision and Robotics Seminar

    Date:
    21
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 18:00-19:00
    Title: Learning on Pointclouds for 3D Scene Understanding
    Lecturer: Or Litany
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk i'll be covering several works in the topic of 3D deep learning on ... Read more In this talk i'll be covering several works in the topic of 3D deep learning on pointclouds for scene understanding tasks. First, I'll describe VoteNet (ICCV 2019, best paper nomination): a method for object detection from 3D pointclouds input, inspired by the classical generalized Hough voting technique. I'll then explain how we integrated image information into the voting scheme to further boost 3D detection (ImVoteNet, CVPR 2020). In the second part of my talk I'll describe recent studies focusing on reducing supervision for 3D scene understanding tasks, including PointContrast -- a self-supervised representation learning framework for 3D pointclods (ECCV 2020). Our findings in PointContrast are extremely encouraging: using a unified triplet of architecture, source dataset, and contrastive loss for pre-training, we achieve improvement over recent best results in segmentation and detection across 6 different benchmarks for indoor and outdoor, real and synthetic datasets -- demonstrating that the learned representation can generalize across domains.
    Close details
    Abstract: ... Read more
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    20
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 16:30-18:00
    Title: An infinitesimal variant of Guo-Jacquet trace formulae and its comparison
    Lecturer: Li Huajie
    Organizer: Faculty of Mathematics and Computer Science
    Details: This talk is based on my thesis supervised by P.-H. Chaudouard. The conjecture o ... Read more This talk is based on my thesis supervised by P.-H. Chaudouard. The conjecture of Guo-Jacquet is a promising generalization to higher dimensions of Waldspurger’s well-known theorem on the relation between toric periods and central values of automorphic L-functions for GL(2). However, we are faced with divergent integrals when applying the relative trace formula approach. In this talk, we study an infinitesimal variant of this problem. Concretely, we establish global and local trace formulae for infinitesimal symmetric spaces of Guo-Jacquet. To compare regular semi-simple terms, we present the weighted fundamental lemma and certain identities between Fourier transforms of local weighted orbital integrals
    Close details

    Foundations of Computer Science Colloquium

    Date:
    18
    Monday
    January
    2021
    Lecture / Seminar
    Time: 16:00-17:30
    Title: Mechanism Design for a Matching Platform: On Multi-Dimensional Gains from Trade Maximization
    Lecturer: Kira Goldner
    Organizer: Faculty of Mathematics and Computer Science
    Details: Consider the algorithm designer for a two-sided market with buyers and sellers, ... Read more Consider the algorithm designer for a two-sided market with buyers and sellers, such as Airbnb or Uber. The platform's goal is to design a mechanism that optimizes the "gains from trade": the aggregate gain in the market due to the trades (e.g. in items, labor) that have occurred. In this talk, I will present the first guarantees on gains from trade for a market that contains a multi-dimensional buyer, who is interested in multiple different kinds of items, and single-dimensional sellers, who each are only interested in selling the single item that they own. We present a logarithmic approximation to the optimal gains from trade in a parameter of the market. We also provide a logarithmic approximation in the number of sellers to the constrained-optimal gains from trade: optimal subject to standard truthfulness constraints. Our techniques come from a variety of domains: including online algorithms and revenue maximization. Joint work with Yang Cai, Steven Ma, and Mingfei Zhao
    Close details

    Vision and Robotics Seminar

    Date:
    14
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Next generation localization microscopy - or - how and why to ruin a perfectly good microscope
    Lecturer: Yoav Shechtman
    Organizer: Faculty of Mathematics and Computer Science
    Details: In localization microscopy, the positions of individual nanoscale point emitters ... Read more In localization microscopy, the positions of individual nanoscale point emitters (e.g. fluorescent molecules) are determined at high precision from their point-spread functions (PSFs). This enables highly precise single/multiple-particle-tracking, as well as super-resolution microscopy, namely single molecule localization microscopy (SMLM). To obtain 3D localization, we employ PSF engineering – namely, we physically modify the standard PSF of the microscope, to encode the depth position of the emitter.
    Close details
    Abstract: In localization microscopy, the positions of individual nanoscale point emitters ... Read more In localization microscopy, the positions of individual nanoscale point emitters (e.g. fluorescent molecules) are determined at high precision from their point-spread functions (PSFs). This enables highly precise single/multiple-particle-tracking, as well as super-resolution microscopy, namely single molecule localization microscopy (SMLM). To obtain 3D localization, we employ PSF engineering – namely, we physically modify the standard PSF of the microscope, to encode the depth position of the emitter. In this talk I will describe how this method enables unprecedented capabilities in localization microscopy
    Close abstract

    Faculty Seminar

    Date:
    14
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 11:00-12:30
    Title: Next Generation Programming with Program Synthesis
    Lecturer: Hila Peleg
    Organizer: Faculty of Mathematics and Computer Science
    Details: Program synthesis is the problem of generating a program to satisfy a specificat ... Read more Program synthesis is the problem of generating a program to satisfy a specification of user intent. Since these specifications are usually partial, this means searching a space of candidate programs for one that exhibits the desired behavior. The lion's share of the work on program synthesis focuses on new ways to perform the search, but hardly any of this research effort has found its way into the hands of users. We wish to use synthesis to augment the programming process, leveraging both optimized search algorithms and concepts that are part of the programmer's life such as code review and read-eval-print loops (REPL). This talk describes three synthesis-based techniques that bring program synthesis into the development workflow. A major concern in designing for the user is that it can put the interface of the synthesizer at odds with state of the art synthesis techniques. Synthesis is, at best, a computationally hard problem, and any changes made to make the tool more usable can interfere with the synthesizer and its internals. We therefore demonstrate the process of bringing synthesis theory into practice when tool design also requires an algorithm re-design. https://weizmann.zoom.us/j/97604026721?pwd=SVZnd3NsQXBqNHFhOU04bTFNQVRhQT09
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    13
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Low degree cohomologies of congruence groups
    Lecturer: Binyong Sun
    Organizer: Faculty of Mathematics and Computer Science
    Details: We determine certain low degree cohomologies of locally symmetric spaces, using ... Read more We determine certain low degree cohomologies of locally symmetric spaces, using representation theory. Basic theory of continuous cohomologies will be reviewed. This is a joint work with Jian-Shu Li
    Close details

    Machine Learning and Statistics Seminar

    Date:
    13
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 15:15-16:45
    Title: Estimation of Manifolds from Point Clouds: Building Models from Data
    Lecturer: Barak Sober
    Organizer: Faculty of Mathematics and Computer Science
    Details: A common observation in data-driven applications is that high dimensional data h ... Read more A common observation in data-driven applications is that high dimensional data has a low intrinsic dimension, at least locally. Thus, when one wishes to work with data that is not governed by a clear set of equations, but still wishes to perform statistical or other scientific analysis, an optional model is the assumption of an underlying manifold from which the data was sampled. This manifold, however, is not given explicitly but we can obtain samples of it (i.e., the individual data points). In this talk, we will consider the mathematical problem of estimating manifolds from a finite set of samples, possibly recorded with noise. Using a Manifold Moving Least-Squares approach we provide an approximant manifold with high approximation order in both the Hausdorff sense as well as in the Riemannian metric (i.e., a nearly isometry). In the case of bounded noise, we present an algorithm that is guaranteed to converge in probability when the number of samples tends to infinity. The motivation for this work is based on the analysis of the evolution of shapes through time (e.g., handwriting or primates' teeth) and we will show how this framework can be utilized to answer scientific questions in paleontology and archaeology.
    Close details

    Vision and Robotics Seminar

    Date:
    07
    Thursday
    January
    2021
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Decoding visual experience from brain activity
    Lecturer: Guy Gaziv
    Organizer: Faculty of Mathematics and Computer Science
    Details: Deep Learning introduced powerful research tools for studying visual representat ... Read more Deep Learning introduced powerful research tools for studying visual representation in the human brain. Here, we harnessed those tools for two branches of research: 1. The primary branch focuses on brain decoding: reconstructing and semantically classifying observed natural images from novel (unknown) fMRI brain recordings. This is a very difficult task due to the scarce supervised “paired” training examples (images with their corresponding fMRI recordings) that are available, even in the largest image-fMRI datasets. We present a self-supervised deep learning approach that overcomes this barrier. This is obtained by enriching the scarce paired training data with additional easily accessible “unpaired” data from both domains (i.e., images without fMRI, and fMRI without images). Our approach achieves state-of-the-art results in image reconstruction from fMRI responses, as well as unprecedented large-scale (1000-way) semantic classification of never-before-seen classes. 2. The secondary branch of research focuses on face representation in the human brain. We studied whether the unique structure of the face-space geometry, which is defined by pairwise similarities in activation patterns to different face images, constitutes a critical aspect in face perception. To test this, we compared the pairwise similarity between responses to face images of human-brain and of artificial Deep Convolutional Neural Networks (DCNN) that achieve human-level face recognition performance. Our results revealed a stark match between neural and intermediate DCNN layers' face-spaces. Our findings support the importance of face-space geometry in enabling face perception as well as a pictorial function of high-order face-selective regions of the human visual cortex. https://weizmann.zoom.us/j/94895425759?pwd=RTh3VkMyamJOay96N3hDcWg0eFpqUT09
    Close details
    Abstract: Deep Learning introduced powerful research tools for studying visual representat ... Read more Deep Learning introduced powerful research tools for studying visual representation in the human brain. Here, we harnessed those tools for two branches of research: 1. The primary branch focuses on brain decoding: reconstructing and semantically classifying observed natural images from novel (unknown) fMRI brain recordings. This is a very difficult task due to the scarce supervised "paired" training examples (images with their corresponding fMRI recordings) that are available, even in the largest image-fMRI datasets. We present a self-supervised deep learning approach that overcomes this barrier. This is obtained by enriching the scarce paired training data with additional easily accessible "unpaired" data from both domains (i.e., images without fMRI, and fMRI without images). Our approach achieves state-of-the-art results in image reconstruction from fMRI responses, as well as unprecedented large-scale (1000-way) semantic classification of never-before-seen classes. 2. The secondary branch of research focuses on face representation in the human brain. We studied whether the unique structure of the face-space geometry, which is defined by pairwise similarities in activation patterns to different face images, constitutes a critical aspect in face perception. To test this, we compared the pairwise similarity between responses to face images of human-brain and of artificial Deep Convolutional Neural Networks (DCNN) that achieve human-level face recognition performance. Our results revealed a stark match between neural and intermediate DCNN layers' face-spaces. Our findings support the importance of face-space geometry in enabling face perception as well as a pictorial function of high-order face-selective regions of the human visual cortex. https://weizmann.zoom.us/j/94895425759?pwd=RTh3VkMyamJOay96N3hDcWg0eFpqUT09
    Close abstract

    Superalgebra Theory and Representations Seminar

    Date:
    06
    Wednesday
    January
    2021
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Grothendieck rings for queer Lie superalgebras
    Lecturer: Shifra Reif
    Organizer: Faculty of Mathematics and Computer Science
    Details: The abstract can be found at the following seminar website:
    Abstract: The abstract can be found at the following seminar website: https://www.math. ... Read more The abstract can be found at the following seminar website: https://www.math.bgu.ac.il/~entova/STARS.html
    Close abstract

    Vision and Robotics Seminar

    Date:
    31
    Thursday
    December
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Demystifying Unsupervised Image Translation through the Lens of Attribute Disentanglement
    Lecturer: Yedid Hoshen
    Organizer: Faculty of Mathematics and Computer Science
    Details: Recent approaches for unsupervised image translation are strongly reliant on gen ... Read more Recent approaches for unsupervised image translation are strongly reliant on generative adversarial training and ad-hoc architectural locality constraints. Despite their appealing results, it can be easily observed that the learned class and content representations are entangled which often hurts the translation performance. We analyse this task under the framework of image disentanglement into meaningful attributes. We first analyse the simpler setting, where the domain of the image and its other attributes are independent. By information arguments, we present a non-adversarial approach (LORD) that carefully designed an information bottleneck for class-content disentanglement. Our approach brings attention to several interesting and poorly explored phenomena, particularly the beneficial inductive biases of latent optimization and conditional generators, and it outperforms the top adversarial and non-adversarial class-content disentanglement methods (e,g, DrNet and MLVAE). By further information constraints, we extend our approach to the standard unsupervised image translation task where the unknown image properties are dependent on the domain. Our full approach surpasses the top unsupervised image translation methods (e.g. FUNIT and StartGAN-v2).
    Close details
    Abstract: Recent approaches for unsupervised image translation are strongly reliant on gen ... Read more Recent approaches for unsupervised image translation are strongly reliant on generative adversarial training and ad-hoc architectural locality constraints. Despite their appealing results, it can be easily observed that the learned class and content representations are entangled which often hurts the translation performance. We analyse this task under the framework of image disentanglement into meaningful attributes. We first analyse the simpler setting, where the domain of the image and its other attributes are independent. By information arguments, we present a non-adversarial approach (LORD) that carefully designed an information bottleneck for class-content disentanglement. Our approach brings attention to several interesting and poorly explored phenomena, particularly the beneficial inductive biases of latent optimization and conditional generators, and it outperforms the top adversarial and non-adversarial class-content disentanglement methods (e,g, DrNet and MLVAE). By further information constraints, we extend our approach to the standard unsupervised image translation task where the unknown image properties are dependent on the domain. Our full approach surpasses the top unsupervised image translation methods (e.g. FUNIT and StartGAN-v2).
    Close abstract

    Superalgebra Theory and Representations Seminar

    Date:
    30
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Pi-systems and closed systems in symmetrizable Kac-Moody algebras
    Lecturer: Krishanu Roy
    Organizer: Faculty of Mathematics and Computer Science
    Details: Dynkin introduced the notion of pi-systems in the process of classifying semi-si ... Read more Dynkin introduced the notion of pi-systems in the process of classifying semi-simple subalgebras of complex semi-simple
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    23
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Around the classification of semisimple algebraic supergroups
    Lecturer: Alexander Sherman
    Organizer: Faculty of Mathematics and Computer Science
    Details: We will discuss the classification of algebraic supergroups G for which their re ... Read more We will discuss the classification of algebraic supergroups G for which their representation category Rep(G) is semisimple (working over an algebraically closed field of characteristic zero). The statement roughly says that OSp(1|2n) is the only 'truly super' algebraic supergroup with this property. We will discuss different proofs and related ideas, with the goal of understanding in some ways how non-semisimplicity expresses itself in a supergroup.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    23
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Harmonic Analysis on GLn over Finite Fields
    Lecturer: Shamgar Gurevich
    Organizer: Faculty of Mathematics and Computer Science
    Details: There are many formulas that express interesting properties of a finite group G ... Read more There are many formulas that express interesting properties of a finite group G in terms of sums over its characters. For estimating these sums, one of the most salient quantities to understand is the character ratio Trace(p(g)) / dim(p), for an irreducible representation p of G and an element g of G. For example, Diaconis and Shahshahani stated a formula of the mentioned type for analyzing certain random walks on G. Recently, we discovered that for classical groups G over finite fields there is a natural invariant of representations that provides strong information on the character ratio. We call this invariant rank. Rank suggests a new organization of representations based on the very few "small"
    Close details
    Abstract: ... Read more
    Close abstract

    Machine Learning and Statistics Seminar

    Date:
    23
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Computational Barriers in Continuous Optimization: Two Complexity Theories and One Tale of Symmetry
    Lecturer: Yossi Arjevani
    Organizer: Faculty of Mathematics and Computer Science
    Details: Since the modern formulation of mathematical optimization, researchers in the fi ... Read more Since the modern formulation of mathematical optimization, researchers in the field have repeatedly expanded and re-defined the realm of tractable optimization problems. This endeavor has culminated in the well-known class of convex optimization problems with applications in a wide range of scientific fields. In this talk, I will present the traditional oracle-based complexity model, which has dominated (unstructured) continuous optimization for the past 40 years, and highlight some of its successes and failures in predicting the hardness of convex optimization problems. I will then introduce a novel structural-based model aimed at addressing major oracle-based complexity gaps. The new approach is intimately tied with approximation theory, and is proven to be particularly advantageous for characterizing the complexity of optimization methods in machine learning. Studies in recent years have indicated that the realm of tractable optimization problems may be expanded once more -- this time into that of the nonconvex world. In the second part of the talk, I will present a novel symmetry-based approach to study nonconvex optimization landscapes, and use it to address optimization-related aspects of neural networks. The approach employs techniques, new to the field, from symmetry breaking and representation theory, and carries important implications for one’s ability to argue about statistical generalization through local curvature. https://weizmann.zoom.us/j/93958126972?pwd=STg0b0JMUFI5eS9rVEd0QXpRY2V2UT09
    Close details
    Abstract: ... Read more
    Close abstract

    Machine Learning and Statistics Seminar

    Date:
    22
    Tuesday
    December
    2020
    Lecture / Seminar
    Time: 15:00-16:00
    Title: On the injectivity and (in)stability of invariant encoding
    Lecturer: Nadav Dym
    Organizer: Faculty of Mathematics and Computer Science
    Details: Quotient spaces are a natural mathematical tool to describe a variety of algorit ... Read more Quotient spaces are a natural mathematical tool to describe a variety of algorithmic problems in computer vision and related fields, where different objects are to be compared while their natural symmetries are to be ignored. The computational complications arising from these symmetries can be alleviated by mapping the given object to features invariant to the object’s symmetries. But can this be done without losing information? In this talk we will discuss this question for two different problems: (a) Neural networks for point clouds: The natural symmetries of 3D point clouds are permutations and rigid motions. We will describe our recent work which provides a general framework for constructing such invariant networks to be lossless, in the sense that they can approximate any continuous invariant function. As a result, invariant networks such as Tensor Field Networks are shown to have universal approximation power. (b) Phase retrieval: Phase retrieval is the problem of retrieving a signal, up to global phase, from linear measurements whose phase is lost. The phase ambiguity here can be considered as the symmetries of the problem, and the phaseless linear measurements as the invariants. We will discuss results on the injectivity and stability of phase retrieval, and particularly our results connecting phase retrieval stability to measures of graph connectivity. To conclude, we will point out some insights which can be obtained from viewing these two different problems in the same framework.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    22
    Tuesday
    December
    2020
    Lecture / Seminar
    Time: 15:00-16:00
    Title: On the injectivity and (in)stability of invariant encoding
    Lecturer: Nadav Dym
    Organizer: Faculty of Mathematics and Computer Science
    Details: Quotient spaces are a natural mathematical tool to describe a variety of algorit ... Read more Quotient spaces are a natural mathematical tool to describe a variety of algorithmic problems in computer vision and related fields, where different objects are to be compared while their natural symmetries are to be ignored. The computational complications arising from these symmetries can be alleviated by mapping the given object to features invariant to the object’s symmetries. But can this be done without losing information? In this talk we will discuss this question for two different problems: (a) Neural networks for point clouds: The natural symmetries of 3D point clouds are permutations and rigid motions. We will describe our recent work which provides a general framework for constructing such invariant networks to be lossless, in the sense that they can approximate any continuous invariant function. As a result, invariant networks such as Tensor Field Networks are shown to have universal approximation power. (b) Phase retrieval: Phase retrieval is the problem of retrieving a signal, up to global phase, from linear measurements whose phase is lost. The phase ambiguity here can be considered as the symmetries of the problem, and the phaseless linear measurements as the invariants. We will discuss results on the injectivity and stability of phase retrieval, and particularly our results connecting phase retrieval stability to measures of graph connectivity. To conclude, we will point out some insights which can be obtained from viewing these two different problems in the same framework.
    Close details

    Vision and Robotics Seminar

    Date:
    17
    Thursday
    December
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: A causal view of compositional zero-shot recognition
    Lecturer: Yuval Atzmon
    Organizer: Faculty of Mathematics and Computer Science
    Details: People easily recognize new visual categories that are new combinations of known ... Read more People easily recognize new visual categories that are new combinations of known components. This compositional generalization capacity is critical for learning in real-world domains like vision and language because the long tail of new combinations dominates the distribution. Unfortunately, learning systems struggle with compositional generalization because they often build on features that are correlated with class labels even if they are not "essential" for the class. This leads to consistent misclassification of samples from a new distribution, like new combinations of known components. In this talk I will present our work on compositional zero-shot recognition: I will describe a causal approach that asks "which intervention caused the image?"
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    16
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 16:30-18:30
    Title: Theta representations and their wavefront sets
    Lecturer: Fan Gao
    Organizer: Faculty of Mathematics and Computer Science
    Details: For a linear algebraic group, a theta representation is just a character of the group

    Superalgebra Theory and Representations Seminar

    Date:
    16
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 10:00-11:15
    Title: A semisimple extension of the Takiff superalgebra
    Lecturer: Kevin Coulembier
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk, I will give an overview of the (finite dimensional) representation ... Read more In this talk, I will give an overview of the (finite dimensional) representation theory of a particular semisimple Lie superalgebra. In particular, I will explain character formulas, extension groups, block decompositions, invariant theory and Koszulity.
    Close details

    Vision and Robotics Seminar

    Date:
    10
    Thursday
    December
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: GANs: Origins, Effective Training, and Steering
    Lecturer: Tomer Michaeli
    Organizer: Faculty of Mathematics and Computer Science
    Details: Since their introduction by Goodfellow et al. in 2014, generative adversarial mo ... Read more Since their introduction by Goodfellow et al. in 2014, generative adversarial models seem to have completely transformed Computer Vision and Graphics. In this talk I will address three questions: (1) What did we do before the GAN era (and was it really that different)? (2) Is the way we train GANs in line with the theory (and can we do it better)? (3) How is information about object transformations encoded in a pre-trained generator? I will start by showing that Contrastive Divergence (CD) learning (Hinton ‘02), the most widely used method for learning distributions before GANs, is in fact also an adversarial procedure. This settles a long standing debate regarding the objective that this method actually optimizes, which arose due to an unjustified approximation in the original derivation. Our observation explains CD’s great empirical success. Going back to GANs, I will challenge the common practice for stabilizing training using spectral-normalization. Although theoretically motivated by the Wasserstein GAN formulation, I will show that this heuristic works for different reasons and can be significantly improved upon. Our improved approach leads to state-of-the-art results in many common tasks, including super-resolution and image-to-image-translation. Finally, I will address the task of revealing meaningful directions in the latent space of a pre-trained GAN. I will show that such directions can be computed in closed form directly from the generator's weights, without the need of any training or optimization as done in existing works. I will particularly discuss nonlinear trajectories that have natural endpoints and allow controlling whether one transformation is allowed to come on the expense of another (e.g. zoom-in with or without allowing translation to keep the object centered). * These are joint works with Omer Yair, Idan Kligvasser, Nurit Spingarn, and Ron Banner. https://weizmann.zoom.us/j/96344142954?pwd=RlhsVk10cWlMMm5WdGhtenRvZGQyQT09
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    09
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Indecomposable summands in tensor products
    Lecturer: Thorsten Heidersdorf
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will report on some progress to understand indecomposable summands in tensor p ... Read more I will report on some progress to understand indecomposable summands in tensor products of irreducible representations of gl(m|n). I will focus on the gl(m|2) -case (m
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    09
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Indecomposable summands in tensor products
    Organizer: Faculty of Mathematics and Computer Science

    Superalgebra Theory and Representations Seminar

    Date:
    02
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Finite-dimensional representation theory of the queer Lie superalgebra q(n)
    Lecturer: Nikolay Grantcharov
    Organizer: Faculty of Mathematics and Computer Science
    Details: We will first describe some classical theory of the queer Lie superalgebras q(n) ... Read more We will first describe some classical theory of the queer Lie superalgebras q(n), such as Clifford modules, (generic) character formula for the irreducible representations, and classification of the blocks of finite-dimensional representations. Then we will focus our attention to q(3) and provide an explicit description of the Ext-quivers of the blocks. A proof of a ``virtual'' BGG reciprocity for q(n), which then gives the radical filtrations of indecomposable projective objects for q(3), will be provided.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    02
    Wednesday
    December
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Selfdual cuspidal representations of GL(r,D) and distinction by an inner involution
    Lecturer: Vincent Secherre
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let n be a positive integer, F be a non-Archimedean locally compact field of odd ... Read more Let n be a positive integer, F be a non-Archimedean locally compact field of odd residue characteristic p and G be an inner form of GL(2n,F). This is a group of the form GL(r,D) for a positive integer r and division F-algebra D of reduced degree d such that rd=2n. Let K be a quadratic extension of F in the algebra of matrices of size r with coefficients in D, and H be its centralizer in G. We study selfdual cuspidal representations of G and their distinction by H, that is, the existence of a nonzero H-invariant linear form on such representations, from the viewpoint of type theory. When F has characteristic 0, we characterize distinction by H for cuspidal representations of G in terms of their Langlands parameter, proving in this case a conjecture by Prasad and Takloo-Bighash.
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    25
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Tensor categories and the superworld
    Lecturer: Inna Entova-Aizenbud
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will give a series of introductory talks on tensor categories and how they giv ... Read more I will give a series of introductory talks on tensor categories and how they give a natural setting for everything "super", such as Lie superalgebras and supergroups.No knowledge of category theory will be assumed - I will explain the categorical terms along the way.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    25
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: : Equations in arithmetic groups and Model Theory
    Lecturer: Nir Avni
    Organizer: Faculty of Mathematics and Computer Science
    Details: I'll talk about a new dichotomy between arithmetic groups of rank one and arithm ... Read more I'll talk about a new dichotomy between arithmetic groups of rank one and arithmetic groups of rank bigger than one. Namely, whereas the category of definable sets in free groups or surface groups is simple, the category of definable sets in many higher rank arithmetic groups is as bad as it gets---it is equivalent to the category of definable sets over the natural numbers. One consequence of this phenomenon is that if G is such a higher rank arithmetic group, then there is an axiom---a first order statement---that holds for G but doesn't hold for any other finitely generated group.
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    18
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Tensor categories and the superworld
    Lecturer: Inna Entova-Aizenbud
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will give a series of introductory talks on tensor categories and how they giv ... Read more I will give a series of introductory talks on tensor categories and how they give a natural setting for everything "super", such as Lie superalgebras and supergroups.
    Close details
    Abstract: ... Read more
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    18
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: A local trace formula for the local Gan-Gross-Prasad conjecture for special orthogonal groups
    Lecturer: Zhilin Luo
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The local Gan-Gross-Prasad conjecture studies the restriction and branching prob ... Read more The local Gan-Gross-Prasad conjecture studies the restriction and branching problems for representations of classical and metaplectic groups. In this talk, I will talk about my proof for the tempered part of the local Gan-Gross-Prasad conjecture (multiplicity one in Vogan packets) for special orthogonal groups over any local fields of characteristic zero, which combines the work of Waldspurger (for the tempered part of the conjecture for special orthogonal groups over $p$-adic fields) and Beuzart-Plessis (for the tempered part of the conjecture for unitary groups over real field) in a non-trivial way. In the proof, an indispensable result which is also of independent interest is a formula expressing the regular nilpotent germs of quasi-split reductive Lie algebras over any local fields of characteristic zero via endoscopic invariants, which was previously proved by Shelstad over $p$-adic fields. We also relate the formula with the Kostant's sections. HTTPS://WEIZMANN.ZOOM.US/J/98304397425
    Close abstract

    Superalgebra Theory and Representations Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: דיווחים
    Lecturer: Maria Gorelik
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series representations of affine symmetric spaces of G = SO(p
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science
    Details: מחלות הסרטן גובות מחיר ממיליוני אנשים מדי ש ... Read more מחלות הסרטן גובות מחיר ממיליוני אנשים מדי שנה. המחקר בתחום פועל בכמה חזיתות: מניעה, אבחון וטיפול מותאם-אישית. תמיכה בחקר הסרטן מעניקה את מתנת החיים לרבים.
    Close details
    Abstract: B. Gross and D. Prasad _rst formulated their famous conjectures about the restr ... Read more B. Gross and D. Prasad _rst formulated their famous conjectures about the restriction of representations of discrete series representations in the original paper Discrete series of an orthogonal group G = SOn when restricted to an orthogonal subgroup G0 = SOn
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series representations of affine symmetric spaces of G = SO(p
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science
    Details: אנחנו, וכל הסובב אותנו, עשויים מאבק כוכבים. ח ... Read more אנחנו, וכל הסובב אותנו, עשויים מאבק כוכבים. חקר מסתרי היקום, הכולל את החלקיקים הזעירים ביותר ואת הגלקסיות העצומות, מחפש תשובות בנוגע למרחב, לזמן ולחיים עצמם – כאן ובמקומות אחרים.
    Close details
    Abstract: B. Gross and D. Prasad _rst formulated their famous conjectures about the restr ... Read more B. Gross and D. Prasad _rst formulated their famous conjectures about the restriction of representations of discrete series representations in the original paper Discrete series of an orthogonal group G = SOn when restricted to an orthogonal subgroup G0 = SOn
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series representations of affine symmetric spaces of G = SO(p
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: From analysis of factors affecting the sea, land, and air on Earth, to the devel ... Read more From analysis of factors affecting the sea, land, and air on Earth, to the development of food security strategies for the Earth’s inhabitants, environmental research aims to secure the health of our planet for generations to come.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series representations of affine symmetric spaces of G = SO(p
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: The immune system plays a critical role in many medical conditions. Groundbreaki ... Read more The immune system plays a critical role in many medical conditions. Groundbreaking immunology research is leading to new strategies for the prevention, diagnosis, and treatment of a wide range of related medical conditions.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series representations of affine symmetric spaces of G = SO(p
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science
    Details: Science is shaping the future, through novel, rationally designed materials that ... Read more Science is shaping the future, through novel, rationally designed materials that form the basis of everything from bio-molecular sensors to quantum circuits. These discoveries will lead to critical and exciting advances in medicine, transportation, communication, and entertainment.
    Close details
    Abstract: B. Gross and D. Prasad _rst formulated their famous conjectures about the restr ... Read more B. Gross and D. Prasad _rst formulated their famous conjectures about the restriction of representations of discrete series representations in the original paper Discrete series of an orthogonal group G = SOn when restricted to an orthogonal subgroup G0 = SOn
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: The decomposition of discrete series
    Lecturer: Birgit Speh
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: Learning how to use the information we create is essential for making technologi ... Read more Learning how to use the information we create is essential for making technological, medical, and scientific developments. Artificial intelligence tools and approaches are enhancing the way we gather, store, secure and analyze big data for the benefit of humanity.
    Close details

    Superalgebra Theory and Representations Seminar

    Date:
    04
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 18:30-19:45
    Title: Superalgebra Theory and Representations Seminar
    Lecturer: Maria Gorelik
    Organizer: Faculty of Mathematics and Computer Science

    Superalgebra Theory and Representations Seminar

    Date:
    04
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 18:00-19:45
    Title: Superalgebra Theory and Representations Seminar
    Organizer: Faculty of Mathematics and Computer Science

    Algebraic Geometry and Representation Theory Seminar

    Date:
    04
    Wednesday
    November
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Double descent for classical groups
    Lecturer: David Soudry
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: We consider the generalized doubling integrals of Cai, Friedberg, Ginzburg and K ... Read more We consider the generalized doubling integrals of Cai, Friedberg, Ginzburg and Kaplan. These generalize the doubling method of Piatetski-Shapiro and Rallis and represent the standard L-function for pairs of irreducible, automorphic, cuspidal representations pi - on a (split) classical group G, and au - onGL(n). The representation pi need not have any particular model (such as a Whittaker model, or a Bessel model). These integrals suggest an explicit descent map (an inverse to Langlands functorial lift) from GL(n) to G(appropriate G). I will show that a certain Fourier coefficient applied to a residual Eisenstein series, induced from a Speh representation, corresponding to a self-dual au, is equal to the direct sum of irreducible cuspidal representations sigma otimes sigma', on G x G , where sigma runs over all irreducible cuspidal representations, which lift to au (sigma' is the complex conjugate of an outer conjugation of sigma). This is a joint work with David Ginzburg. https://weizmann.zoom.us/j/98304397425
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    28
    Wednesday
    October
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Briefs
    Lecturer: Lei Zhang
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: In this talk, we will discuss the theory of twisted automorphic descents, which ... Read more In this talk, we will discuss the theory of twisted automorphic descents, which is an extension of the automorphic descent of Ginzburg-Rallis-Soudry.
    Close details
    Abstract: ... Read more
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    28
    Wednesday
    October
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Bessel-Fourier coefficients and Twisted Automorphic Descent
    Lecturer: Lei Zhang
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: In this talk, we will discuss the theory of twisted automorphic descents, which ... Read more In this talk, we will discuss the theory of twisted automorphic descents, which is an extension of the automorphic descent of Ginzburg-Rallis-Soudry. The main goal is to construct cuspidate automorphic modules in the generic global Arthur packets by using Bessel-Fourier coefficients of automorphic representations. Moreover, we will discuss some applications and problems related Bessel-Fourier coefficients. This is a joint work with Dihua Jiang.
    Close details
    Abstract: ... Read more
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    21
    Wednesday
    October
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: On multiplicativity of gamma-factors and Fourier transforms via Braverman-Kazhdan program
    Lecturer: Freydoon Shahidi
    Organizer: Faculty of Mathematics and Computer Science
    Details: This is a joint work with my student William Sokurski. Braverman-Kazhdan/Ngo pro ... Read more This is a joint work with my student William Sokurski. Braverman-Kazhdan/Ngo program aims to generalize the work of Godement-Jacquet/Tate from GL(n) to an arbitrary reductive group G and a finite dimensional representation r of its L-group. We briefly review the general concepts of the method, including Renner's construction of reductive r-monoids, objects generalizing the space of n x n matrices in the case of GL(n), concluding with the example of symmetric power representations of GL(2,C).
    Close details
    Abstract: This is a joint work with my student William Sokurski. Braverman-Kazhdan/Ngo pro ... Read more This is a joint work with my student William Sokurski. Braverman-Kazhdan/Ngo program aims to generalize the work of Godement-Jacquet/Tate from GL(n) to an arbitrary reductive group G and a finite dimensional representation r of its L-group. We briefly review the general concepts of the method, including Renner's construction of reductive r-monoids, objects generalizing the space of n x n matrices in the case of GL(n), concluding with the example of symmetric power representations of GL(2,C). We then define a space of r-Schwartz functions interms of the restriction of the conjectural r-Fourier transform to the space of smooth functions of compact support in G, as in the work of Braverman and Kazhdan. Multiplicativity which states the equality of gamma factors for the parabolically induced and inducing data, follows from a natural commutativity of corresponding Fourier transforms on G and the Levi subgroup L, sharing the fixed maximal torus defining the monoid, with a generalized Harish-Chandra transform. We finally present a candidate for the Fourier transform attached to the symmetric cube of GL(2,C) as a fiber integration over the compact-inducing data for tamely ramified supercuspidals of GL(2) constructed by Howe and Bushnell-Kutzko, as predicted by Ngo. This last result is part of Sokurski's thesis which can be extended to all the odd symmetric powers. https://weizmann.zoom.us/j/98304397425
    Close abstract

    Algebraic Geometry and Representation Theory Seminar

    Date:
    14
    Wednesday
    October
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Relative endoscopy for certain symmetric spaces
    Lecturer: Spencer Leslie
    Organizer: Faculty of Mathematics and Computer Science
    Details: Motivated by problems arising from the study of certain relative trace formulas, ... Read more Motivated by problems arising from the study of certain relative trace formulas, I discuss a notion of endoscopy in a relative setting. The main example is that of unitary Friedberg-Jacquet periods, which are related to special cycles in certain unitary Shimura varieties. After introducing the endoscopic symmetric spaces in this case, I will sketch the proof of the fundamental lemma. https://weizmann.zoom.us/j/98304397425
    Close details

    Seminar in Geometry and Topology

    Date:
    12
    Monday
    October
    2020
    Lecture / Seminar
    Time: 09:00-11:00
    Title: A proof of A. Gabrielov's rank Theorem
    Lecturer: Andre Belotto da Silva
    Organizer: Faculty of Mathematics and Computer Science,Faculty of Mathematics and Computer Science
    Details: This talk concerns Gabrielov's rank Theorem, a fundamental result in local compl ... Read more This talk concerns Gabrielov's rank Theorem, a fundamental result in local complex and real-analytic geometry, proved in the 1970's. Contrasting with the algebraic case, it is not in general true that the analytic rank of an analytic map (that is, the dimension of the analytic-Zariski closer of its image) is equal to the generic rank of the map (that is, the generic dimension of its image). This phenomena is behind several pathological examples in local real-analytic geometry. Gabrielov's rank Theorem provides a formal condition for the equality to hold https://weizmann.zoom.us/j/9183356684?pwd=QlN4T1VtcXc4QXdXNEYxUlQ4dmR5dz09
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    30
    Wednesday
    September
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: A-packets for quasisplit GSp(2n) and GSO(2n) over a p-adic field
    Lecturer: Bin Xu
    Organizer: Faculty of Mathematics and Computer Science
    Details: Arthur (1989) conjectured that the discrete spectrum of automorphic representati ... Read more Arthur (1989) conjectured that the discrete spectrum of automorphic representations of a connected reductive group over a number field can be decomposed into global A-packets, in terms of which he also conjectured a multiplicity formula. Arthur (2013) proved his conjectures for symplectic and orthogonal groups, in which case the global A-packets are parametrized by self-dual automorphic representations of general linear groups. In this talk, I will give a construction of the local A-packets for general symplectic and general even orthogonal groups in the nonarchimedean case. This is based on our earlier works in the tempered case, and it follows a construction by Moeglin for symplectic and orthogonal groups. ZOOM: HTTPS://WEIZMANN.ZOOM.US/J/98304397425
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    16
    Wednesday
    September
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: On the subring of special cycles on orthogonal Shimura varieties
    Lecturer: Stephen Kudla
    Organizer: Faculty of Mathematics and Computer Science
    Details: By old results with Millson, the generating series for the cohomology classes of ... Read more By old results with Millson, the generating series for the cohomology classes of special cycles on orthogonal Shimura varieties over a totally real field are Hilbert-Siegel modular forms. These forms arise via theta series. Using this result and the Siegel-Weil formula, we show that the products in the subring of cohomology generated by the special cycles are controlled by the Fourier coefficients of triple pullbacks of certain Siegel-Eisenstein series. As a consequence, there are comparison isomorphisms between special subrings for different Shimura varieties. In the case in which the signature of the quadratic space V is (m,2) at an even number d_ of archimedean places, the comparison gives a `combinatorial model' for the special cycle ring in terms of the associated totally positive definite space.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    02
    Wednesday
    September
    2020
    Lecture / Seminar
    Time: 16:30-17:35
    Title: Automorphy in string scattering and small representations
    Lecturer: Axel Kleinschmidt
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will review how automorphic representations arise in the calculation of string ... Read more I will review how automorphic representations arise in the calculation of string theory scattering amplitudes. As follows from the work of Green, Miller, Vanhove, Pioline and others, the automorphic representations are associated with split real groups of a certain exceptional family. In the cases that are well understood, these representation have very small Gelfand-Kirillov dimension. Their Fourier expansion can be calculated using different methods and confirms physical expectation on the wavefront set. In work with Gourevitch, Gustafsson, Persson and Sahi, the method of Whittaker pairs was employed to systematize this analysis. I will also comment on the cases that are less well understood in physics and that appear to go beyond the standard notion of automorphic representations since the usual Z-finiteness condition is violated.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    26
    Wednesday
    August
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: Ellipticity and discrete series
    Lecturer: Bernhard Kroetz
    Organizer: Faculty of Mathematics and Computer Science
    Details: We explain by elementary means why the existence of a discrete series representa ... Read more We explain by elementary means why the existence of a discrete series representation of a real reductive group G implies the existence of a compact Cartan subgroup of G. The presented approach has the potential to generalize to real spherical spaces. The talk will be based on https://arxiv.org/pdf/2007.15312.pdf ZOOM: HTTPS://WEIZMANN.ZOOM.US/J/98304397425
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    August
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: On automorphic descent from GL(7) to G2
    Lecturer: Baiying Liu
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk, I will introduce the functorial descent from cuspidal automorphic ... Read more In this talk, I will introduce the functorial descent from cuspidal automorphic representations pi of GL7(A) with L^S(s, pi, wedge^3) having a pole at s=1 to the split exceptional group G2(A), using Fourier coefficients associated to two nilpotent orbits of E7. We show that one descent module is generic, and under mild assumptions on the unramified components of pi, it is cuspidal and having pi as a weak functorial lift of each irreducible summand. However, we show that the other descent module supports not only the non-degenerate Whittaker integral on G2(A), but also every degenerate Whittaker integral. Thus it is generic, but not cuspidal. This is a new phenomenon, compared to the theory of functorial descent for classical and GSpin groups. This work is joint with Joseph Hundley.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    12
    Wednesday
    August
    2020
    Lecture / Seminar
    Time: 16:30-17:20
    Title: On automorphic descent from GL(7) to G2
    Organizer: Faculty of Mathematics and Computer Science

    Seminar in Geometry and Topology

    Date:
    03
    Monday
    August
    2020
    Lecture / Seminar
    Time: 16:00-17:30
    Title: GOOD COMPACTIFICATION THEOREM FOR (C*)^n
    Lecturer: Askold Khovanskii
    Organizer: Faculty of Mathematics and Computer Science
    Details:

    Vision and Robotics Seminar

    Date:
    30
    Thursday
    July
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Semantic Pyramid for Image Generation
    Lecturer: Assaf Shocher
    Organizer: Faculty of Mathematics and Computer Science
    Details: We present a novel GAN-based model that utilizes the space of deep features lear ... Read more We present a novel GAN-based model that utilizes the space of deep features learned by a pre-trained classification model. Inspired by classical image pyramid representations, we construct our model as a Semantic Generation Pyramid - a hierarchical framework which leverages the continuum of semantic information encapsulated in such deep features
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    29
    Wednesday
    July
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: On the distinction of Harish-Chandra modules and its Ext-analogues.
    Lecturer: Wen-Wei Li
    Organizer: Faculty of Mathematics and Computer Science

    Seminar in Geometry and Topology

    Date:
    27
    Monday
    July
    2020
    Lecture / Seminar
    Time: 16:00-17:30
    Title: The role of Puiseux characteristics in the local Poincaré problem
    Lecturer: Javier Ribon
    Organizer: Faculty of Mathematics and Computer Science
    Details: The usual Poincare problem consists in determining whether or not there exist up ... Read more The usual Poincare problem consists in determining whether or not there exist upper bounds for the degree of an invariant algebraic curve of a foliation of the complex projective plane in terms of the degree of the foliation. We are interested in the local analogue,
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    22
    Wednesday
    July
    2020
    Lecture / Seminar
    Time: 16:30-15:30
    Title: A new theta correspondence
    Lecturer: Solomon Friedberg
    Organizer: Faculty of Mathematics and Computer Science
    Details: The classical theta correspondence establishes a relationship between automorphi ... Read more The classical theta correspondence establishes a relationship between automorphic representations on special orthogonal groups and automorphic representations on symplectic groups or their double covers. This correspondence is achieved by using as integral kernel a theta series that is constructed from the Weil representation. In this talk I will briefly survey earlier work on (local and global, classical and other) theta correspondences and then present an extension of the classical theta correspondence to higher degree metaplectic covers. The key issue here is that for higher degree covers there is no analogue of the Weil representation (or even a minimal representation), so additional ingredients are needed. Joint work with David Ginzburg. ZOOM: HTTPS://WEIZMANN.ZOOM.US/J/98304397425
    Close details

    Seminar in Geometry and Topology

    Date:
    20
    Monday
    July
    2020
    Lecture / Seminar
    Time: 16:00-17:30
    Title: Limit cycle enumeration for random vector fields
    Lecturer: Erik Lundberg
    Organizer: Faculty of Mathematics and Computer Science
    Details: The second part of Hilbert's sixteenth problem asks for a study of the number an ... Read more The second part of Hilbert's sixteenth problem asks for a study of the number and relative positions of the limit cycles of an ODE system associated to a planar vector field with polynomial component functions.
    Close details

    Seminar in Geometry and Topology

    Date:
    16
    Thursday
    July
    2020
    Lecture / Seminar
    Time: 17:00-18:30
    Title: Infinitesimal Center Problem on zero cycles and the composition conjecture
    Lecturer: Pavao Mardešić
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will present a joint work with A. Alvarez, J.L. Bravo and C. Christopher.

    Algebraic Geometry and Representation Theory Seminar

    Date:
    15
    Wednesday
    July
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Transfer of characters under the Howe duality correspondence
    Lecturer: Wee Teck Gan
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let Sp(W) x O(V) be a dual reductive pair. If

    Algebraic Geometry and Representation Theory Seminar

    Date:
    08
    Wednesday
    July
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Nilpotent orbits associated to distinguished representations of reductive groups.
    Lecturer: Dmitry Gourevitch
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let G be a reductive group over a local field F of characteristic zero, and H be ... Read more Let G be a reductive group over a local field F of characteristic zero, and H be a spherical subgroup. An irreducible representation of G is said to be distinguished by H if it has an H-invariant continuous linear functional. The study of distinguished representations is of much current interest, because of their relation to the Plancherel measure on G/H and to periods of automorphic forms. While a complete classification seems to be out of reach, we established simple micro-local necessary conditions for distinction. The conditions are formulated in terms of the nilpotent orbits associated to the representation, in the spirit of the orbit method. Our results are strongest for Archimedean F. In this case, Rossmann showed that for any irreducible Casselman-Wallach representation, the Zariski closure of the wave-front set is the closure of a unique nilpotent complex orbit. We have shown that the restriction of this orbit to the complexified Lie algebra of H includes zero. We apply this result to symmetric pairs, branching problems, and parabolic induction. We also have a twisted version for the case when
    Close details

    Seminar in Geometry and Topology

    Date:
    29
    Monday
    June
    2020
    Lecture / Seminar
    Time: 09:15-10:45
    Title: Asymptotic positivity
    Lecturer: Yanir Rubinstein
    Organizer: Faculty of Mathematics and Computer Science
    Details: A general theme in geometry is the classification of algebraic/differential geom ... Read more A general theme in geometry is the classification of algebraic/differential geometric structures which satisfy a positivity property. I will describe an
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    24
    Wednesday
    June
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Restriction for general linear groups: the local non-tempered Gan-Gross-Prasad conjecture
    Lecturer: Kei Yuen Chan
    Organizer: Faculty of Mathematics and Computer Science
    Details: Recently, Gan-Gross-Prasad formulated new restriction problems for the non-tempe ... Read more Recently, Gan-Gross-Prasad formulated new restriction problems for the non-tempered representations of classical groups. In this talk, I shall explain a proof of a conjecture on general linear groups over non-Archimedean fields. The main ingredients of the proof include a use of filtration on parabolically induced representations when restricted to the mirabolic subgroups, and realizing the product with a Speh representation as a functor. The proof also uses a result of Lapid- Mínguez on the irreducibility of a product of representations. If time permits, we shall also discuss generalizations to Bessel and Fourier-Jacobi models and towards Ext-branching laws. Zoom link:
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    17
    Wednesday
    June
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Singularity properties of convolutions of algebraic morphisms and probabilistic Waring type problems
    Lecturer: Yotam Hendel
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let G be a connected algebraic group.

    Vision and Robotics Seminar

    Date:
    11
    Thursday
    June
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Explorable Super Resolution
    Lecturer: Yuval Bahat
    Organizer: Faculty of Mathematics and Computer Science
    Details: Single image super resolution (SR) has seen major performance leaps in recent ye ... Read more Single image super resolution (SR) has seen major performance leaps in recent years. However, existing methods do not allow exploring the infinitely many plausible reconstructions that might have given rise to the observed low-resolution (LR) image. These different explanations to the LR image may dramatically vary in their textures and fine details, and may often encode completely different semantic information. In this work, we introduce the task of explorable super resolution. We propose a framework comprising a graphical user interface with a neural network backend, allowing editing the SR output so as to explore the abundance of plausible HR explanations to the LR input. At the heart of our method is a novel module that can wrap any existing SR network, analytically guaranteeing that its SR outputs would precisely match the LR input, when downsampled. Besides its importance in our setting, this module is guaranteed to decrease the reconstruction error of any SR network it wraps, and can be used to cope with blur kernels that are different from the one the network was trained for. We illustrate our approach in a variety of use cases, ranging from medical imaging and forensics, to graphics. Zoom link:
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    10
    Wednesday
    June
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: Unitarity, Eisenstein series, and Arthur's conjectures
    Lecturer: Stephen Miller
    Organizer: Faculty of Mathematics and Computer Science
    Details: Jim Arthur has conjectured the existence of some exotic "unipotent" representati ... Read more Jim Arthur has conjectured the existence of some exotic "unipotent" representations of real reductive Lie groups, which are expected to form building blocks of the unitary dual. Though falling short of a full classification of the unitary dual itself, Arthur's conjectures touch on the essence of some of the most difficult questions concerning unitarity. In another direction, automorphic realizations of these representations are expected to have delicate arithmetic properties. However, Arthur's unipotent representations are hard to identify, much show are unitary. I'll present a status report, including the unitary of the "Langlands element" Arthur describes directly (in joint work with Joe Hundley), and the full identification the unipotent representations for exceptional real groups (joint work with Jeff Adams, Marc van Leeuwen, Annegret Paul, and David Vogan). Zoom meeting:
    Close details

    Vision and Robotics Seminar

    Date:
    04
    Thursday
    June
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: A Method For Removing Water From Underwater Images
    Lecturer: Tali Treibitz
    Organizer: Faculty of Mathematics and Computer Science
    Details: Robust recovery of lost colors in underwater images remains a challenging proble ... Read more Robust recovery of lost colors in underwater images remains a challenging problem. We recently showed that this was partly due to the prevalent use of an atmospheric image formation model for underwater images and proposed a physically accurate model. The revised model showed: 1) the attenuation coefficient of the signal is not uniform across the scene but depends on object range and reflectance, 2) the coefficient governing the increase in backscatter with distance differs from the signal attenuation coefficient. Here, we present the first method that recovers color with our revised model, using RGBD images. The Sea-thru method estimates backscatter using the dark pixels and their known range information. Then, it uses an estimate of the spatially varying illuminant to obtain the range-dependent attenuation coefficient. Using more than 1,100 images from two optically different water bodies, which we make available, we show that our method with the revised model outperforms those using the atmospheric model. Consistent removal of water will open up large underwater datasets to powerful computer vision and machine learning algorithms, creating exciting opportunities for the future of underwater exploration and conservation. (Paper published in CVPR 19). Zoom link: https://weizmann.zoom.us/j/99608045732
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    03
    Wednesday
    June
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: A relative de Rham theorem for Nash Submersions
    Lecturer: Shachar Carmeli
    Organizer: Faculty of Mathematics and Computer Science
    Details: For a Nash manifold X and a Nash vector bundle E on X, one can form the topologi ... Read more For a Nash manifold X and a Nash vector bundle E on X, one can form the topological vector space of Schwartz sections of E, i.e. the smooth sections which decay fast along with all derivatives. It was shown by Aizenbud and Gourevitch, and independently by Luca Prelli, that for a Nash manifold X, th complex of Schwartz sections of the de Rham complex of X has cohomologies isomorphic to the compactly supported cohomologies of X. In my talk I will present a work in progress, joint with Avraham Aizenbud, to generalize this result to the relative case, replacing the Nash manifold M with a Nash submersion f:M--
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    03
    Wednesday
    June
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: A relative de Rham theorem for Nash Submersions
    Organizer: Faculty of Mathematics and Computer Science
    Details: For a Nash manifold X and a Nash vector bundle E on X, one can form the topologi ... Read more For a Nash manifold X and a Nash vector bundle E on X, one can form the topological vector space of Schwartz sections of E, i.e. the smooth sections which decay fast along with all derivatives. It was shown by Aizenbud and Gourevitch, and independently by Luca Prelli, that for a Nash manifold X, th complex of Schwartz sections of the de Rham complex of X has cohomologies isomorphic to the compactly supported cohomologies of X. In my talk I will present a work in progress, joint with Avraham Aizenbud, to generalize this result to the relative case, replacing the Nash manifold M with a Nash submersion f:M--
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    27
    Wednesday
    May
    2020
    Lecture / Seminar
    Time: 16:30-17:30
    Title: Beilinson-Bernstein localization via wonderful asymptotics.
    Lecturer: Iordan Ganev
    Organizer: Faculty of Mathematics and Computer Science
    Details: We explain how a doubled version of the Beilinson-Bernstein localization functor ... Read more We explain how a doubled version of the Beilinson-Bernstein localization functor can be understood using the geometry of the wonderful compactification of a group. Specifically, bimodules for the Lie algebra give rise to monodromic D-modules on the horocycle space, and to filtered D-modules on the group that respect a certain matrix coefficients filtration. These two categories of D-modules are related via an associated graded construction in a way compatible with localization, Verdier specialization, the Vinberg semigroup, and additional structures. This talk is based on joint work with David Ben-Zvi. Zoom meeting: https://weizmann.zoom.us/j/98304397425
    Close details

    Faculty Seminar

    Date:
    25
    Monday
    May
    2020
    Lecture / Seminar
    Time: 17:30-18:30
    Title: Randomness extraction and amplification in the quantum world
    Lecturer: Rotem Arnon-Friedman
    Organizer: Faculty of Mathematics and Computer Science
    Details: Randomness is an essential resource in computer science. In most applications pe ... Read more Randomness is an essential resource in computer science. In most applications perfect, and sometimes private, randomness is needed, while it is not even clear that such a resource exists. It is well known that the tools of classical computer science do not allow us to create perfect randomness from a single weak source of randomness -
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    20
    Wednesday
    May
    2020
    Lecture / Seminar
    Time: 16:30-18:00
    Title: Isolation of the cuspidal spectrum and application to the Gan-Gross-Prasad conjecture for unitary groups.
    Lecturer: Raphael Beuzart-Plessis
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk, I will explain a new way to construct smooth convolution operators ... Read more In this talk, I will explain a new way to construct smooth convolution operators on adelic groups that isolate (certain) cuspidal representations from the rest of the automorphic spectrum. Then, I will explain an application of this construction to the global Gan-Gross-Prasad conjecture for unitary groups. This is joint work with Yifeng Liu, Wei Zhang and Xinwen Zhu. Zoom meeting https://weizmann.zoom.us/j/98304397425
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    05
    Thursday
    March
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: DOUBLE SEMINAR
    Location: Jacob Ziskind Building
    Lecturer: Emanuel Milman (Technion)
    Organizer: Faculty of Mathematics and Computer Science
    Details: Speaker #1: Emanuel Milman (Technion) Title: Functional Inequalities on sub-R ... Read more Speaker #1: Emanuel Milman (Technion) Title: Functional Inequalities on sub-Riemannian manifolds via QCD Abstract:We are interested in obtaining Poincar’e and log-Sobolev inequalities on domains in sub-Riemannian manifolds (equipped with their natural sub-Riemannian metric and volume measure). It is well-known that strictly sub-Riemannian manifolds do not satisfy any type of Curvature-Dimension condition CD(K,N), introduced by Lott-Sturm-Villani some 15 years ago, so we must follow a different path. We show that while ideal (strictly) sub-Riemannian manifolds do not satisfy any type of CD condition, they do satisfy a quasi-convex relaxation thereof, which we name QCD(Q,K,N). As a consequence, these spaces satisfy numerous functional inequalities with exactly the same quantitative dependence (up to a factor of Q) as their CD counterparts. We achieve this by extending the localization paradigm to completely general interpolation inequalities, and a one-dimensional comparison of QCD densities with their "CD upper envelope". We thus obtain the best known quantitative estimates for (say) the L^p-Poincar’e and log-Sobolev inequalities on domains in the ideal sub-Riemannian setting, which in particular are independent of the topological dimension. For instance, the classical Li-Yau / Zhong-Yang spectral-gap estimate holds on all Heisenberg groups of arbitrary dimension up to a factor of 4. No prior knowledge will be assumed, and we will (hopefully) explain all of the above notions during the talk. Speaker #2: Tal Orenshtein (TU Berlin) Title: Rough walks in random environment Abstract. In this talk we shall review scaling limits for random walks in random environment lifted to the rough path space to the enhanced Brownian motion. Except for the immediate application to SDEs, this adds some new information on the structure of the limiting path. Time permitted, we shall elaborate on the tools to tackle these problems. Based on joint works with Olga Lopusanschi, with Jean-Dominique Deuschel and Nicolas Perkowski and with Johaness Bäumler, Noam Berger and Martin Slowik.
    Close details

    Language of Evolution and Evolution of Language

    Date:
    23
    Sunday
    February
    2020
    -
    24
    Monday
    February
    2020
    Conference
    Time: 08:00
    Location: David Lopatie Conference Centre

    Foundations of Computer Science Colloquium

    Date:
    03
    Monday
    February
    2020
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Local Proofs Approaching the Witness Length
    Location: Jacob Ziskind Building
    Lecturer: Ron Rothblum
    Organizer: Faculty of Mathematics and Computer Science
    Details: Interactive oracle proofs (IOPs) are a hybrid between interactive proofs and PCP ... Read more Interactive oracle proofs (IOPs) are a hybrid between interactive proofs and PCPs. In an IOP the prover is allowed to interact with a verifier (like in an interactive proof) by sending relatively long messages to the verifier, who in turn is only allowed to query a few of the bits that were sent (like in a PCP). For any NP relation for which membership can be decided in polynomial-time and bounded polynomial space (e.g., SAT, Hamiltonicity, Clique, Vertex-Cover, etc.) and for any constant gamma
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    30
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Anti-concentration of inner products
    Location: Jacob Ziskind Building
    Lecturer: Amir Yehudayoff
    Organizer: Faculty of Mathematics and Computer Science
    Details: The plan is to discuss anti-concentration of the inner product between two indep ... Read more The plan is to discuss anti-concentration of the inner product between two independent random vectors in Euclidean space. We shall go over some background and applications, and see some proofs.
    Close details

    Vision and Robotics Seminar

    Date:
    30
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Handling the Unknown with Non-Rigid Geometric Invariants
    Location: Jacob Ziskind Building
    Lecturer: Oshri Halimi
    Organizer: Faculty of Mathematics and Computer Science
    Details: My goal is to demonstrate how geometric priors can replace the requirement for a ... Read more My goal is to demonstrate how geometric priors can replace the requirement for annotated 3D data. In this context I'll present two of my works. First, I will present a deep learning framework that estimates dense correspondence between articulated 3D shapes without using any ground truth labeling which I presented as an oral presentation at CVPR 2019: "Unsupervised Learning of Dense Shape Correspondence". We demonstrated that our method is applicable to full and partial 3D models as well as to realistic scans. The problem of incomplete 3D data can be encountered also in many other different scenarios. One such interesting problem of great practical importance is shape completion, to which I'll dedicate the second part of my lecture. It is common to encounter situations where there is a considerable distinction between the scanned 3D model and the final rendered one. The distinction can be attributed to occlusions or directional view of the target, when the scanning device is localized in space. I will demonstrate that geometric priors can guide learning algorithms in the task of 3D model completion from partial observation. To this end, I'll present our recent work: "The Whole is Greater than the Sum of its Non-Rigid Parts". This famous declaration of Aristotle was adopted to explain human perception by the Gestalt psychology school of thought in the twentieth century. Here, we claim that observing part of an object which was previously acquired as a whole, one could deal with both partial matching and shape completion in a holistic manner. More specifically, given the geometry of a full, articulated object in a given pose, as well as a partial scan of the same object in a different pose, we address the problem of matching the part to the whole while simultaneously reconstructing the new pose from its partial observation. Our approach is data-driven, and takes the form of a Siamese autoencoder without the requirement of a consistent vertex labeling at inference time
    Close details

    Computer Science Seminar

    Date:
    29
    Wednesday
    January
    2020
    Lecture / Seminar
    Time: 13:00-14:00
    Title: Interaction is necessary for distributed learning with privacy or communication constraints
    Location: Jacob Ziskind Building
    Lecturer: Yuval Dagan
    Organizer: Faculty of Mathematics and Computer Science
    Details: Local differential privacy (LDP) is a model where users send privatized data to ... Read more Local differential privacy (LDP) is a model where users send privatized data to an untrusted central server whose goal it to solve some data analysis task. In the non-interactive version of this model the protocol consists of a single round in which a server sends requests to all users then receives their responses. This version is deployed in industry due to its practical advantages and has attracted significant research interest. Our main result is an exponential lower bound on the number of samples necessary to solve the standard task of learning a large-margin linear separator in the non-interactive LDP model. Via a standard reduction this lower bound implies an exponential lower bound for stochastic convex optimization and specifically, for learning linear models with a convex, Lipschitz and smooth loss. These results answer the questions posed in citep{SmithTU17,DanielyF18}. Our lower bound relies on a new technique for constructing pairs of distributions with nearly matching moments but whose supports can be nearly separated by a large margin hyperplane. These lower bounds also hold in the model where communication from each user is limited and follow from a lower bound on learning using non-adaptive emph{statistical queries}. Link: https://arxiv.org/abs/1911.04014
    Close details

    Faculty Seminar

    Date:
    26
    Sunday
    January
    2020
    Lecture / Seminar
    Time: 11:15-13:00
    Title: On Parameterized Analysis and the Disjoint Paths Problem
    Location: Jacob Ziskind Building
    Lecturer: Meirav Zehavi
    Organizer: Faculty of Mathematics and Computer Science
    Details: Parameterized Anaylsis leads both to deeper understanding of intractability resu ... Read more Parameterized Anaylsis leads both to deeper understanding of intractability results and to practical solutions for many NP-hard problems. Informally speaking, Parameterized Analysis is a mathematical paradigm to answer the following fundamental question: What makes an NP-hard problem hard? Specifically, how do different parameters (being formal quantifications of structure) of an NP-hard problem relate to its inherent difficulty? Can we exploit these relations algorithmically, and to which extent? Over the past three decades, Parameterized Analysis has grown to be a mature field of outstandingly broad scope with significant impact from both theoretical and practical perspectives on computation. In this talk, I will first give a brief introduction of the field of Parameterized Analysis. Then, I will discuss some recent work in this field, where I mainly address (i) problems at the core of this field, rooted at Graph Minors and Graph Modification Problems, and (ii) applications of tools developed in (i) in particular, and of parameterized analysis in general, to Computational Geometry and Computational Social Choice. Additionally, I will zoom into a specific result, namely, the first single-exponential time parameterized algorithm for the Disjoint Paths problem on planar graphs. An efficient algorithm for the Disjoint Paths problem in general, and on "almost planar" graphs in particular, is a critical part in the quest for the establishment of an Efficient Graph Minors Theory. As the Graph Minors Theory is the origin of Parameterized Analysis and ubiquitous in the design of parameterized algorithms, making this theory efficient is a holy grail in the field.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    23
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Metric entropies, L_1 transportation, and competitive analysis
    Location: Jacob Ziskind Building
    Lecturer: James Lee
    Organizer: Faculty of Mathematics and Computer Science
    Details: The MTS problem (Borodon, Linial, and Saks 1992) is a general model for the anal ... Read more The MTS problem (Borodon, Linial, and Saks 1992) is a general model for the analysis of algorithms that optimize in the presence of information arriving over time, where the state space is equipped with a metric. I will discuss a relatively new approach to this area, where both the algorithm and method of analysis are derived canonically from a choice of a "regularizer" on the probability simplex. The regularizer can be interpreted as a Riemannian structure on the simplex, and then the algorithm is simply a gradient flow. Defining the regularizer as an appropriate "noisy" multiscale metric entropy (similar to, e.g., the Talagrand gamma_1 functional) yields the best-known competitive ratio for every metric space. For ultrametrics (aka, "HST metrics"), this achieves the conjectured bound, but the algorithm falls short of resolving the MTS conjecture for many other spaces, including subsets of the real line.
    Close details

    Vision and Robotics Seminar

    Date:
    23
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Learning to Sample
    Location: Jacob Ziskind Building
    Lecturer: Oren Dovrat
    Organizer: Faculty of Mathematics and Computer Science
    Details: Processing large point clouds is a challenging task. Therefore, the data is ofte ... Read more Processing large point clouds is a challenging task. Therefore, the data is often sampled to a size that can be processed more easily. The question is how to sample the data? A popular sampling technique is Farthest Point Sampling (FPS). However, FPS is agnostic to a downstream application (classification, retrieval, etc.). The underlying assumption seems to be that minimizing the farthest point distance, as done by FPS, is a good proxy to other objective functions. We show that it is better to learn how to sample. To do that, we propose a deep network to simplify 3D point clouds. The network, termed S-NET, takes a point cloud and produces a smaller point cloud that is optimized for a particular task. The simplified point cloud is not guaranteed to be a subset of the original point cloud. Therefore, we match it to a subset of the original points in a post-processing step. We contrast our approach with FPS by experimenting on two standard data sets and show significantly better results for a variety of applications. Our code is publicly available.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    21
    Tuesday
    January
    2020
    Lecture / Seminar
    Time: 11:15-12:30
    Title: A bridge between p-adic and quantum group representations via Whittaker coinvariants.
    Location: Jacob Ziskind Building
    Lecturer: Valentin Buciumas
    Organizer: Faculty of Mathematics and Computer Science
    Details: Unramified principal series representations of p-adic GL(r) and its metaplectic ... Read more Unramified principal series representations of p-adic GL(r) and its metaplectic covers are important in the theory of automorphic forms. I will present a method of relating the Whittaker coinvariants of such a representation with representations of quantum affine gl_n. This involves using a Schur-Weyl duality result due to Chari and Pressley and it allows us to compute the dimension of the Whittaker model of every irreducible smooth representation with Iwahori fixed vectors. If time permits I will explain a conjectured version of this result for the symplectic group Sp(2r) which involves quantum symmetric pairs.
    Close details

    Foundations of Computer Science Colloquium

    Date:
    20
    Monday
    January
    2020
    Lecture / Seminar
    Time: 11:15-13:00
    Title: A New Analysis of Differential Privacy’s Generalization Guarantees
    Location: Jacob Ziskind Building
    Lecturer: Katrina Ligett
    Organizer: Faculty of Mathematics and Computer Science
    Details: Many data analysis pipelines are adaptive: the choice of which analysis to run n ... Read more Many data analysis pipelines are adaptive: the choice of which analysis to run next depends on the outcome of previous analyses. Common examples include variable selection for regression problems and hyper-parameter optimization in large-scale machine learning problems: in both cases, common practice involves repeatedly evaluating a series of models on the same dataset. Unfortunately, this kind of adaptive re-use of data invalidates many traditional methods of avoiding over-fitting and false discovery, and has been blamed in part for the recent flood of non-reproducible findings in the empirical sciences. An exciting line of work beginning with Dwork et al. 2015 establishes the first formal model and first algorithmic results providing a general approach to mitigating the harms of adaptivity, via a connection to the notion of differential privacy. Unfortunately, until now, those results were primarily of information theoretic interest, only beating out the simple approach of gathering fresh data for every computation ("sample-splitting") at the scale of many millions of datapoints. In this work, we give a new proof of the transfer theorem that any mechanism for answering adaptively chosen statistical queries that is differentially private and sample-accurate is also accurate out-of-sample. Our new proof is elementary and gives structural insights that we expect will be useful elsewhere. We show: 1) that differential privacy ensures that the expectation of any query on the conditional distribution on datasets induced by the transcript of the interaction is close to its expectation on the data distribution, and 2) sample accuracy on its own ensures that any query answer produced by the mechanism is close to the expectation of the query on the conditional distribution. This second claim follows from a thought experiment in which we imagine that the dataset is resampled from the conditional distribution after the mechanism has committed to its answers. The transfer theorem then follows by summing these two bounds, and in particular, avoids the "monitor argument" used to derive high probability bounds in prior work. An upshot of our new proof technique is that the concrete bounds we obtain are substantially better than the best previously known bounds, even though the improvements are in the constants, rather than the asymptotics (which are known to be tight). As we show, our new bounds outperform the naive "sample-splitting" baseline at dramatically smaller dataset sizes compared to the previous state of the art, bringing techniques from this literature closer to practicality. Joint work with: Christopher Jung, Seth Neel, Aaron Roth, Saeed Sharifi-Malvajerdi (UPenn), and Moshe Shenfeld (HUJI). This work appeared at ITCS 2020.
    Close details

    Computer Science Seminar

    Date:
    19
    Sunday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:00
    Title: Learning vs. Verifying
    Location: Jacob Ziskind Building
    Lecturer: Jonathan Shafer
    Organizer: Faculty of Mathematics and Computer Science
    Details: This talk will address

    Faculty Seminar

    Date:
    19
    Sunday
    January
    2020
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Verification of distributed protocols using decidable logics
    Location: Jacob Ziskind Building
    Lecturer: Oded Padon
    Organizer: Faculty of Mathematics and Computer Science
    Details: Formal verification of infinite-state systems, and distributed systems in partic ... Read more Formal verification of infinite-state systems, and distributed systems in particular, is a long standing research goal. I will describe a series of works that develop a methodology for verifying distributed algorithms and systems using decidable logics, employing decomposition, abstraction, and user interaction. This methodology is implemented in an open-source tool, and has resulted in the first mechanized proofs of several important distributed protocols. I will also describe a novel approach to the problem of invariant inference based on a newly formalized duality between reachability and mathematical induction. The duality leads to a primal-dual search algorithm, and a prototype implementation already handles challenging examples that other state-of-the-art techniques cannot handle. I will briefly describe several other works, including a new optimization technique for deep learning computations that achieves significant speedups relative to existing deep learning frameworks.
    Close details

    Vision and Robotics Seminar

    Date:
    16
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 12:15-13:00
    Title: Factor Graph Attention
    Location: Jacob Ziskind Building
    Lecturer: Idan Schwartz
    Organizer: Faculty of Mathematics and Computer Science
    Details: Dialog is an effective way to exchange information, but subtle details and nuanc ... Read more Dialog is an effective way to exchange information, but subtle details and nuances are extremely important. While significant progress has paved a path to address visual dialog with algorithms, details and nuances remain a challenge. Attention mechanisms have demonstrated compelling results to extract details in visual question answering and also provide a convincing framework for visual dialog due to their interpretability and effectiveness. However, the many data utilities that accompany visual dialog challenge existing attention techniques. We address this issue and develop a general attention mechanism for visual dialog which operates on any number of data utilities. To this end, we design a factor graph based attention mechanism which combines any number of utility representations. We illustrate the applicability of the proposed approach on the challenging and recently introduced VisDial datasets, outperforming recent state-of-the-art methods by 1.1% for VisDial0.9 and by 2% for VisDial1.0 on MRR. Our ensemble model improved the MRR score on VisDial1.0 by more than 6%.
    Close details

    Faculty Seminar

    Date:
    16
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 11:00-12:30
    Title: Higher Criticism for discriminating frequency-tables and testing authorship
    Location: Gerhard M.J. Schmidt Lecture Hall
    Lecturer: Alon Kipnis
    Organizer: Faculty of Mathematics and Computer Science
    Details: The Higher Criticism (HC) test is a useful tool for detecting the presence of a ... Read more The Higher Criticism (HC) test is a useful tool for detecting the presence of a signal spread across a vast number of features, especially in the sparse setting when only few features are useful while the rest are pure noise. We adapt the HC test to the two-sample setting of detecting changes between two frequency tables. We apply this adaptation to authorship attribution challenges, where the goal is to identify the author of a document using other documents whose authorship is known. The method is simple yet performs well without handcrafting and tuning. Furthermore, as an inherent side effect, the HC calculation identifies a subset of discriminating words, which allow additional interpretation of the results. Our examples include authorship in the Federalist Papers, machine-generated texts, and the identity of the creator of the Bitcoin. We take two approaches to analyze the success of our method. First, we show that, in practice, the discriminating words identified by the test have low variance across documents belonging to a corpus of homogeneous authorship. We conclude that in testing a new document against the corpus of an author, HC is mostly affected by words characteristic of that author and is relatively unaffected by topic structure. Finally, we analyze the power of the test in discriminating two multinomial distributions under a sparse and weak perturbation model. We show that our test has maximal power in a wide range of the model parameters, even though these parameters are unknown to the user.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    15
    Wednesday
    January
    2020
    Lecture / Seminar
    Time: 11:15-12:30
    Title: The Frobenius functor for symmetric tensor categories in positive characteristic.
    Location: The David Lopatie Hall of Graduate Studies
    Lecturer: Pavel Eringof
    Organizer: Faculty of Mathematics and Computer Science
    Details: An important role in modular representation theory is played by the Frobenius tw ... Read more An important role in modular representation theory is played by the Frobenius twist functor, twisting the k-linear structure of a representation by the Frobenius automorphism F(a)=a^p of the (algebraically closed) ground field k of characteristic p. I will define an analog of this functor for any symmetric tensor category of characteristic p. One of the main new features is that unlike the classical Frobenius twist functor, this functor need not be left or right exact. I will give examples when it is not and describe a replacement of the exactness property. I will also describe applications of this notion to formulating and proving analogs of Deligne's theorem in positive characteristic. This is joint work with V. Ostrik.
    Close details

    Computer Science Seminar

    Date:
    13
    Monday
    January
    2020
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Strong Average-Case Circuit Lower Bounds from Non-trivial Derandomization
    Location: Jacob Ziskind Building
    Lecturer: Lijie Chen
    Organizer: Faculty of Mathematics and Computer Science
    Details: We prove that for all constants a, NQP = NTIME[n^polylog(n)] cannot be (1/2 2^ ... Read more We prove that for all constants a, NQP = NTIME[n^polylog(n)] cannot be (1/2 2^(-log^a n) )-approximated by 2^(log^a n)-size ACC^0 of THR circuits (ACC^0 circuits with a bottom layer of threshold gates). Previously, it was even open whether E^NP can be (1/2 1/sqrt(n))-approximated by AC^0[2] circuits.
    Close details

    Seminar in Geometry and Topology

    Date:
    13
    Monday
    January
    2020
    Lecture / Seminar
    Time: 09:15-10:30
    Title: Newton non-degenerate codimension one foliations and blowing-ups
    Location: Jacob Ziskind Building
    Lecturer: Beatriz Molina Samper
    Organizer: Faculty of Mathematics and Computer Science

    Seminar in Geometry and Topology

    Date:
    13
    Monday
    January
    2020
    Lecture / Seminar
    Time: 09:15-10:30
    Title: Newton non-degenerate codimension one foliations and blowing-ups
    Location: Jacob Ziskind Building
    Lecturer: Beatriz Molina Samper
    Organizer: Faculty of Mathematics and Computer Science

    Faculty Seminar

    Date:
    12
    Sunday
    January
    2020
    Lecture / Seminar
    Time: 11:00-12:45
    Title: Reliability, Equity, and Reproducibility in Modern Machine Learning
    Location: Jacob Ziskind Building
    Lecturer: Yaniv Romano
    Organizer: Faculty of Mathematics and Computer Science
    Details: Modern machine learning algorithms have achieved remarkable performance in a myr ... Read more Modern machine learning algorithms have achieved remarkable performance in a myriad of applications, and are increasingly used to make impactful decisions in the hiring process, criminal sentencing, healthcare diagnostics
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    09
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: The Geometric measure theory of the Brownian path
    Location: Jacob Ziskind Building
    Lecturer: Abel Farkas
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let B denote the range of the Brownian motion in R^d. For a deterministic Borel a

    Vision and Robotics Seminar

    Date:
    09
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Beyond Accuracy: Neural Networks Show Similar Learning Dynamics Across Architectures
    Location: Jacob Ziskind Building
    Lecturer: Daphna Weinshall
    Organizer: Faculty of Mathematics and Computer Science
    Details: One of the unresolved questions in deep learning is the nature of the solutions ... Read more One of the unresolved questions in deep learning is the nature of the solutions that are being discovered. We investigated the collection of solutions reached by the same neural network (NN) architecture, with different random initialization of weights and random mini-batches. These solutions are shown to be rather similar -- more often than not, each train and test example is either classified correctly by all NN instances, or by none at all. Furthermore, all NNs seem to share the same learning dynamics, whereby initially the same train and test examples are correctly recognized by the learned model, followed by other examples that are learned in roughly the same order. When extending the investigation to heterogeneous collections of NN architectures, once again examples are seen to be learned in the same order irrespective of architecture, although the more powerful architecture may continue to learn and thus achieve higher accuracy. Finally, I will discuss cases where this pattern of similarity breaks down, which show that the reported similarity is not an artifact of optimization by gradient descent.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    08
    Wednesday
    January
    2020
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Fundamental limits of modern machine learning and how to get around them
    Location: Jacob Ziskind Building
    Lecturer: Yair Carmon
    Organizer: Faculty of Mathematics and Computer Science
    Details: This talk presents new computational and statistical barriers in machine learnin ... Read more This talk presents new computational and statistical barriers in machine learning, along with the algorithmic developments that they inspire. The computational barriers arise in nonconvex optimization: we prove lower bounds on the (oracle) complexity of finding stationary points using (stochastic) gradient methods, showing that gradient descent is unimprovable for a natural class of problems. We bypass this barrier by designing an algorithm that outperforms gradient descent for a large subclass of problems with high-order smoothness. Our algorithm leverages classical momentum techniques from convex optimization using a "convex until proven guilty" principle that we develop. The statistical barrier is the large amount of data required for adversarially robust learning. In a Gaussian model, we prove that unlabeled data allows us to circumvent an information theoretic gap between robust and standard classification. Our analysis directly leads to a general robust self-training procedure
    Close details

    Faculty Seminar

    Date:
    07
    Tuesday
    January
    2020
    Lecture / Seminar
    Time: 11:00-12:30
    Title: Traces of Class/Cross-Class Structure Pervade Deep Learning Spectra
    Location: Jacob Ziskind Building
    Lecturer: Vardan Papyan
    Organizer: Faculty of Mathematics and Computer Science
    Details: Numerous researchers recently applied empirical spectral analysis to the study o ... Read more Numerous researchers recently applied empirical spectral analysis to the study of modern deep learning classifiers, observing spectral outliers and small but distinct bumps often seen beyond the edge of a "main bulk". This talk presents an important formal class/cross-class structure and shows how it lies at the origin of these visually striking patterns. The structure is shown to permeate the spectra of deepnet features, backpropagated errors, gradients, weights, Fisher Information matrix and Hessian, whether these are considered in the context of an individual layer or the concatenation of them all. The significance of the structure is illustrated by (i) demonstrating empirically that the feature class means separate gradually from the bulk as function of depth and become increasingly more orthogonal, (ii) proposing a correction to KFAC, a well known second-order optimization algorithm for training deepnets, and (ii) proving in the context of multinomial logistic regression that the ratio of outliers to bulk in the spectrum of the Fisher information matrix is predictive of misclassification.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    05
    Sunday
    January
    2020
    Lecture / Seminar
    Time: 09:45-11:00
    Title: Beyond Worst Case In Machine Learning: The Oracle Model
    Location: Jacob Ziskind Building
    Lecturer: Alon Gonen
    Organizer: Faculty of Mathematics and Computer Science
    Details: In recent years there has been an increasing gap between the success of machine ... Read more In recent years there has been an increasing gap between the success of machine learning algorithms and our ability to explain their success theoretically. Namely, many of the problems that are solved to a satisfactory degree of precision are computationally hard in the worst case. Fortunately, there are often reasonable assumptions which help us to get around these worst-case impediments and allow us to rigorously analyze heuristics that are used in practice. In this talk I will advocate a complementary approach, where instead of explicitly characterizing some desired "niceness" properties of the data, we assume access to an optimization oracle that solves a relatively simpler task. This allows us to identify the sources of hardness and extend our theoretical understanding to new domains. Furthermore we will show that seemingly innocents (and arguably justifiable) modifications to the oracle can lead to tractable reductions and even to bypass hardness results. We demonstrate these ideas using the following results: i) An efficient algorithm for non-convex online learning using an optimization oracle. b) A faster boosting algorithm using a "simple" weak learner. iii) An efficient reduction from online to private learning. Joint works with Naman Agarwal, Noga Alon, Elad Hazan, and Shay Moran.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    02
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Edgeworth expansion in Central Limit Theorem
    Location: Jacob Ziskind Building
    Lecturer: Dima Dolgopyat
    Organizer: Faculty of Mathematics and Computer Science
    Details: It is well known that Central Limit Theorem is very effective giving a reasonabl ... Read more It is well known that Central Limit Theorem is very effective giving a reasonable approximation for sums of a quite small number of terms. Edgeworth expansions provide a convenient way to control the error in the Central Limit Theorem. In this talk I will review some recent results on this subject related to the interface between probability and dynamics.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    02
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Edgeworth expansion in Central Limit Theorem
    Location: Jacob Ziskind Building
    Lecturer: Dima Dolgopyat
    Organizer: Faculty of Mathematics and Computer Science
    Details: It is well known that Central Limit Theorem is very effective giving a reasonabl ... Read more It is well known that Central Limit Theorem is very effective giving a reasonable approximation for sums of a quite small number of terms. Edgeworth expansions provide a convenient way to control the error in the Central Limit Theorem. In this talk I will review some recent results on this subject related to the interface between probability and dynamics.
    Close details

    Vision and Robotics Seminar

    Date:
    02
    Thursday
    January
    2020
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Meaning Representation in Natural Language Tasks
    Location: Jacob Ziskind Building
    Lecturer: Gabriel Stanovsky
    Organizer: Faculty of Mathematics and Computer Science
    Details: Recent developments in Natural Language Processing (NLP) allow models to leverag ... Read more Recent developments in Natural Language Processing (NLP) allow models to leverage large, unprecedented amounts of raw text, culminating in impressive performance gains in many of the field"s long-standing challenges, such as machine translation, question answering, or information retrieval. In this talk, I will show that despite these advances, state-of-the-art NLP models often fail to capture crucial aspects of text understanding. Instead, they excel by finding spurious patterns in the data, which lead to biased and brittle performance. For example, machine translation models are prone to translate doctors as men and nurses as women, regardless of context. Following, I will discuss an approach that could help overcome these challenges by explicitly representing the underlying meaning of texts in formal data structures. Finally, I will present robust models that use such explicit representations to effectively identify meaningful patterns in real-world texts, even when training data is scarce.
    Close details

    Faculty Seminar

    Date:
    30
    Monday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: What have we Learned from the Archimedes Palimpsest?
    Location: Gerhard M.J. Schmidt Lecture Hall
    Lecturer: Reviel Netz
    Organizer: Faculty of Mathematics and Computer Science
    Details: The rediscovery of the Archimedes Palimpsest in 1998 was a unique event: one of ... Read more The rediscovery of the Archimedes Palimpsest in 1998 was a unique event: one of the most important authors of antiquity was read afresh, with many transformations made in the established text. Twenty years later, what have we learned from the Archimedes Palimpsest? What changed in our picture of the ancient history of mathematics? Reviel Netz is Pat Suppes Professor of Greek Mathematics and Astronomy at Stanford University. Besides editing the Archimedes Palimpsest, he has published many books and articles on the history of Greek mathematics and on other subjects. His most recent book, "Scale, Space and Canon in Ancient Literary Culture", is now forthcoming from Cambridge University Press. Refreshments will be served at 11:00am.
    Close details

    Seminar in Geometry and Topology

    Date:
    30
    Monday
    December
    2019
    Lecture / Seminar
    Time: 09:15-10:45
    Title: Cross-ratio dynamics on ideal polygons
    Location: Jacob Ziskind Building
    Lecturer: Serge Tabachnikov
    Organizer: Faculty of Mathematics and Computer Science
    Details: Define a relation between labeled ideal polygons in the hyperbolic space by requ ... Read more Define a relation between labeled ideal polygons in the hyperbolic space by requiring that the complex distances (a combination of the distance and the angle) between their
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    26
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: A new proof of the Caffarelli contraction theorem
    Location: Jacob Ziskind Building
    Lecturer: Max Fathi
    Organizer: Faculty of Mathematics and Computer Science
    Details: The Caffarelli contraction theorem states that the Brenier optimal transport map ... Read more The Caffarelli contraction theorem states that the Brenier optimal transport map sending the Gaussian measure onto a uniformly log-concave probability measure is lipschitz. In this talk, I will present a new proof, using entropic regularization and a variational characterization of lipschitz transport maps due to Gozlan and Juillet. Based on joint work with Nathael Gozlan and Maxime Prod'homme.
    Close details
    Abstract: The Caffarelli contraction theorem states that the Brenier optimal transport ma ... Read more The Caffarelli contraction theorem states that the Brenier optimal transport map sending the Gaussian measure onto a uniformly log-concave probability measure is lipschitz. In this talk, I will present a new proof, using entropic regularization and a variational characterization of lipschitz transport maps due to Gozlan and Juillet. Based on joint work with Nathael Gozlan and Maxime Prod'homme.
    Close abstract

    Vision and Robotics Seminar

    Date:
    26
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Pixel Consensus Voting for Panoptic Segmentation
    Location: Jacob Ziskind Building
    Lecturer: Greg Shakhnarovich
    Organizer: Faculty of Mathematics and Computer Science
    Details: I will present a new approach for image parsing, Pixel Consensus Voting (PCV). T ... Read more I will present a new approach for image parsing, Pixel Consensus Voting (PCV). The core of PCV is a framework for instance segmentation based on the Generalized Hough transform. Pixels cast discretized, probabilistic votes for the likely regions that contain instance centroids. At the detected peaks that emerge in the voting heatmap, backprojection is applied to collect pixels and produce instance masks. Unlike a sliding window detector that densely enumerates object proposals, our method detects instances as a result of the consensus among pixel-wise votes. We implement vote aggregation and backprojection using native operators of a convolutional neural network. The discretization of centroid voting reduces the training of instance segmentation to pixel labeling, analogous and complementary to fully convolutional network-style semantic segmentation, leading to an efficient and unified architecture that jointly models things and stuff. We demonstrate the effectiveness of our pipeline on COCO and Cityscapes Panoptic Segmentation and obtain competitive results. This joint work with Haochen Wang (TTIC/CMU), Ruotian Luo (TTIC), Michael Maire (TTIC/Uchicago) received an Innovation Award at the COCO/Mapillary workshop at ICCV 2019.
    Close details

    Faculty Seminar

    Date:
    26
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 11:00-13:00
    Title: Leveraging Programmable Switches for In-network Computing
    Lecturer: Ran Ben Basat
    Organizer: Faculty of Mathematics and Computer Science
    Details: The network line rate is constantly on the rise to support the exploding amounts ... Read more The network line rate is constantly on the rise to support the exploding amounts of data. This means that we have less time to process individual packets, despite a growing demand for better network telemetry. Moreover, CPU speeds are not rising at the same rate as we near the end of Moore's law, making it harder to rely on software computations. Programmable hardware switches are an emerging technology that enables flexible packet processing while optimizing for throughput and latency. In this talk, I will introduce algorithms that leverage programmable switches for accelerating database operations and for improving network telemetry. Switches are orders of magnitude more efficient than traditional hardware accelerators, exist in the datapath, and are ideal for computation offloading. For telemetry, we will discuss how switches can probabilistically encode information across multiple packets to provide fine-grained network visibility with minimal overheads.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    25
    Wednesday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Beyond Worst Case In Machine Learning: The Oracle Model
    Location: Jacob Ziskind Building
    Lecturer: Alon Gonen
    Organizer: Faculty of Mathematics and Computer Science
    Details: In recent years there has been an increasing gap between the success of machine ... Read more In recent years there has been an increasing gap between the success of machine learning algorithms and our ability to explain their success theoretically.
    Close details

    Foundations of Computer Science Colloquium

    Date:
    23
    Monday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Non-Interactive Publicly Verifiable Delegation with Applications to PPAD Hardness
    Location: Jacob Ziskind Building
    Lecturer: Yael Tauman Kalai
    Organizer: Faculty of Mathematics and Computer Science
    Details: We construct a delegation scheme for delegating polynomial time computations. Ou ... Read more We construct a delegation scheme for delegating polynomial time computations. Our scheme is publicly verifiable and non-interactive in the common reference string (CRS) model. The soundness of our scheme is based on an efficiently falsifiable decisional assumption on groups with bilinear maps. Prior to this work, publicly verifiable non-interactive delegation schemes were only known under knowledge assumptions (or in the Random Oracle model) or under non-standard assumptions related to obfuscation or multilinear maps. In addition, our scheme has two desirable features: The proofs are unambiguous, in the sense that it is hard to find two distinct proofs for the same statement, and are updatable in the sense that given a proof for the statement that a Turing machine M transitions from configuration C_0 to C_T in T steps, one can efficiently generate a proof for the statement that M transitions from configuration C_0 to C_{T 1} in T 1 steps. We show that such a delegation scheme implies PPAD hardness, by following a similar approach to that of Choudhuri et al. (STOC2019), who obtained PPAD hardness based on an assumption related to the soundness of the Fiat-Shamir paradigm.
    Close details
    Abstract: We construct a delegation scheme for delegating polynomial time computations. Ou ... Read more We construct a delegation scheme for delegating polynomial time computations. Our scheme is publicly verifiable and non-interactive in the common reference string (CRS) model. The soundness of our scheme is based on an efficiently falsifiable decisional assumption on groups with bilinear maps. Prior to this work, publicly verifiable non-interactive delegation schemes were only known under knowledge assumptions (or in the Random Oracle model) or under non-standard assumptions related to obfuscation or multilinear maps. In addition, our scheme has two desirable features: The proofs are unambiguous, in the sense that it is hard to find two distinct proofs for the same statement, and are updatable in the sense that given a proof for the statement that a Turing machine M transitions from configuration C_0 to C_T in T steps, one can efficiently generate a proof for the statement that M transitions from configuration C_0 to C_{T 1} in T 1 steps. We show that such a delegation scheme implies PPAD hardness, by following a similar approach to that of Choudhuri et al. (STOC2019), who obtained PPAD hardness based on an assumption related to the soundness of the Fiat-Shamir paradigm. This is based on two joint works, both with Omer Paneth and Lisa Yang.
    Close abstract

    Computer Science Seminar

    Date:
    22
    Sunday
    December
    2019
    Lecture / Seminar
    Time: 13:30-14:30
    Title: Nearly Instance-Optimal Mechanisms in Differential Privacy
    Location: Jacob Ziskind Building
    Lecturer: Hilal Asi
    Organizer: Faculty of Mathematics and Computer Science
    Details: We develop differentially private mechanisms that achieve nearly instance-optima ... Read more We develop differentially private mechanisms that achieve nearly instance-optimal losses, achieving lower loss than all appropriately unbiased mechanisms for any possible instance. We show that our mechanisms, with a modest increase in sample size (logarithmic or constant), are instance-optimal for a large family of functions. In contrast to existing mechanisms, which use the global or local sensitivity of the function being estimated, and so are necessarily instance suboptimal, the key to our construction is to use the inverse of the sensitivity. This allows a simple instance-optimal algorithm, and we develop several representative private mechanisms, including for the median and regression problems.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    19
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: DOUBLE SEMINAR
    Location: Jacob Ziskind Building
    Lecturer: Esty Kelman
    Organizer: Faculty of Mathematics and Computer Science
    Details: Esty Kelman [TAU]

    Vision and Robotics Seminar

    Date:
    19
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Hue-Net: Intensity-based Image-to-Image Translation with Differentiable Histogram Loss Functions
    Location: Jacob Ziskind Building
    Lecturer: Tammy Riklin Raviv
    Organizer: Faculty of Mathematics and Computer Science
    Details: In the talk I will present the Hue-Net - a novel Deep Learning framework for Int ... Read more In the talk I will present the Hue-Net - a novel Deep Learning framework for Intensity-based Image-to-Image Translation. The key idea is a new technique we term network augmentation which allows a differentiable construction of intensity histograms from images. We further introduce differentiable representations of (1D) cyclic and joint (2D) histograms and use them for defining loss functions based on cyclic Earth Mover's Distance (EMD) and Mutual Information (MI). While the Hue-Net can be applied to several image-to-image translation tasks, we choose to demonstrate its strength on color transfer problems, where the aim is to paint a source image with the colors of a different target image. Note that the desired output image does not exist and therefore cannot be used for supervised pixel-to-pixel learning. This is accomplished by using the HSV color-space and defining an intensity-based loss that is built on the EMD between the cyclic hue histograms of the output and the target images. To enforce color-free similarity between the source and the output images, we define a semantic-based loss by a differentiable approximation of the MI of these images.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    19
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Robinson-Schensted-Knuth correspondence at the service of p-adic GL_n
    Location: Elaine and Bram Goldsmith Building for Mathematics and Computer Sciences
    Lecturer: Max Gurevich
    Organizer: Faculty of Mathematics and Computer Science
    Details: In a joint work with Erez Lapid we constructed a new class of representations ba ... Read more In a joint work with Erez Lapid we constructed a new class of representations based on applying the RSK algorithm on Zelevinski's multisegments. Those constructions have the potential to be an alternative to the commonly used basis of standard representations. Intriguingly, this class also turned out to categorify a 45-year-old development in invariant theory: The Rota basis of standard bitableaux. I will talk about this classical theme and its relation to representations of p-adic GL_n, as well the expected properties of our new class.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    17
    Tuesday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Jordan properties of automorphism groups of varieties
    Location: Jacob Ziskind Building
    Lecturer: Yuri Zarhin
    Organizer: Faculty of Mathematics and Computer Science
    Details: A classical theorem of Jordan asserts that each finite subgroup of the complex g ... Read more A classical theorem of Jordan asserts that each finite subgroup of the complex general linear group GL(n) is "almost commutative": it contains a commutative normal subgroup with index bounded by an universal constant that depends only on n. We discuss an analogue and variants of this property for the groups of birational (and biregular) automorphisms of complex algebraic varieties, the diffeomorphisms groups of real manifolds and the groups of bimeromorphic (and biholomorphic) automorphisms of compact complex manifolds.
    Close details

    Foundations of Computer Science Colloquium

    Date:
    16
    Monday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Pseudo-random pseudo-distributions
    Location: Jacob Ziskind Building
    Lecturer: Gil Cohen
    Organizer: Faculty of Mathematics and Computer Science
    Details: In this talk, we will discuss a new type of a pseudo-random object called a "pse ... Read more In this talk, we will discuss a new type of a pseudo-random object called a "pseudo-random pseudo-distribution". This object was introduced in the context of the BPL vs. L problem, and I will sketch a space-efficient construction of the latter for read-once branching programs that has near-optimal dependence on the error parameter. The talk is a distillation of a joint work with Mark Braverman and Sumegha Garg (the paper is available online: https://eccc.weizmann.ac.il/report/2017/161/).
    Close details

    Faculty Seminar

    Date:
    15
    Sunday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Representation, inference and design of multicellular systems
    Location: Jacob Ziskind Building
    Lecturer: Mor Nitzan
    Organizer: Faculty of Mathematics and Computer Science
    Details: The past decade has witnessed the emergence of single-cell technologies that mea ... Read more The past decade has witnessed the emergence of single-cell technologies that measure the expression level of genes at a single-cell resolution. These developments have revolutionized our understanding of the rich heterogeneity, structure, and dynamics of cellular populations, by probing the states of millions of cells, and their change under different conditions or over time. However, in standard experiments, information about the spatial context of cells, along with additional layers of information they encode about their location along dynamic processes (e.g. cell cycle or differentiation trajectories), is either lost or not explicitly accessible. This poses a fundamental problem for elucidating collective tissue function and mechanisms of cell-to-cell communication. In this talk I will present computational approaches for addressing these challenges, by learning interpretable representations of structure, context and design principles for multicellular systems from single-cell information. I will first describe how the locations of cells in their tissue of origin and the resulting spatial gene expression can be probabilistically inferred from single-cell information by a generalized optimal-transport optimization framework, that can flexibly incorporate prior biological assumptions or knowledge derived from experiments. Inference in this case is based on an organization principle for spatial gene expression, namely a structural correspondence between distances of cells in expression and physical space, which we hypothesized and supported for different tissues. We used this framework to spatially reconstruct diverse tissues and organisms, including the fly embryo, mammalian intestinal epithelium and cerebellum, and further inferred spatially informative genes. Since cells encode multiple layers of information, in addition to their spatial context, I will also discuss several approaches for the disentanglement of single-cell gene expression into distinct biological processes, based on ideas rooted in random matrix theory and manifold learning. I will finally discuss how these results can be generalized to reveal principles underlying self-organization of cells into multicellular structures, setting the foundation for the computationally-directed design of cell-to-cell interactions optimized for specific tissue structure or function.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    12
    Thursday
    December
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Existence of persistence exponent for Gaussian stationary functions
    Location: Jacob Ziskind Building
    Lecturer: Naomi Feldheim
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let Z(t) be a Gaussian stationary function on the real line, and fix a level L

    Computer Science Seminar

    Date:
    09
    Monday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Clustering: How hard is it to classify data?
    Location: Jacob Ziskind Building
    Lecturer: Karthik C.S
    Organizer: Faculty of Mathematics and Computer Science
    Details: Two popular objectives optimized in clustering algorithms are k-means and k-medi ... Read more Two popular objectives optimized in clustering algorithms are k-means and k-median. The k-means (resp. k-median) problem in the L_p-metric is specified by n points as input and the goal is to classify the input point-set into k clusters such that the k-means (resp. k-median) objective is minimized. The best-known inapproximability factor in literature for these NP-hard problems in L_p-metrics were well-below 1.01. In this talk, we take a significant step to improve the hardness of approximating these problems in various L_p-metrics.
    Close details

    Vision and Robotics Seminar

    Date:
    04
    Wednesday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Across Scales
    Location: Jacob Ziskind Building
    Lecturer: Liad Pollak
    Organizer: Faculty of Mathematics and Computer Science
    Details: When a very fast dynamic event is recorded with a low framerate camera, the resu ... Read more When a very fast dynamic event is recorded with a low framerate camera, the resulting video suffers from severe motion blur (due to exposure time) and motion aliasing (due to low sampling rate in time). True Temporal Super-Resolution (TSR) is more than just Temporal-Interpolation (increasing framerate). It also recovers new high temporal frequencies beyond the temporal nyquist limit of the input video, thus resolving both motion-blur and motion-aliasing. In this work we propose a "Deep Internal Learning" approach for true TSR. We train a video-specific CNN on examples extracted directly from the low-framerate input video. Our method exploits the strong recurrence of small space-time patches inside a single video sequence, both within and across different spatio-temporal scales of the video. We further observe (for the first time) that small space-time patches recur also across-dimensions of the video sequence - i.e., by swapping the spatial and temporal dimensions. In particular, the higher spatial resolution of video frames provides strong examples as to how to increase the temporal resolution of that video. Such internal video-specific examples give rise to strong self-supervision, requiring no data but the input video itself. This results in Zero-Shot Temporal-SR of complex videos, which removes both motion blur and motion aliasing, outperforming previous supervised methods trained on external video datasets. * Joint work with Shai Bagon, Eyal Naor, George Pisha, Michal Irani
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    04
    Wednesday
    December
    2019
    Lecture / Seminar
    Time: 09:30-10:45
    Title: Cubic Fourfolds: Rationality and Derived Categories
    Location: Jacob Ziskind Building
    Lecturer: Howard Nuer
    Organizer: Faculty of Mathematics and Computer Science
    Details: The question of determining if a given algebraic variety is rational is a notori ... Read more The question of determining if a given algebraic variety is rational is a notoriously difficult problem in algebraic geometry, and attempts to solve rationality problems have often produced powerful new techniques. A well-known open rationality problem is the determination of a criterion for when a cubic hypersurface of five-dimensional projective space is rational. After discussing the history of this problem, I will introduce the two conjectural rationality criteria that have been put forth and then discuss a package of tools I have developed with my collaborators to bring these two conjectures together. Our theory of Relative Bridgeland Stability has a number of other beautiful consequences such as a new proof of the integral Hodge Conjecture for Cubic Fourfolds and the construction of full-dimensional families of projective HyperKahler manifolds. Time permitting I'll discuss applications of the theory of relative stability conditions to problems other than cubic fourfolds.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    03
    Tuesday
    December
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: The generalized doubling method: multiplicity one and its applications
    Location: Jacob Ziskind Building
    Lecturer: Eyal Kaplan
    Organizer: Faculty of Mathematics and Computer Science
    Details: The doubling method, first introduced by Piatetski-Shapiro and Rallis in the 80s ... Read more The doubling method, first introduced by Piatetski-Shapiro and Rallis in the 80s, has had numerous applications, e.g. to the theta correspondence and to arithmetic problems. In a series of recent works this method was generalized in several aspects, with an application to functoriality from classical groups to GL(N). One crucial ingredient for the development of the theory is a multiplicity one result, obtained recently in a joint work with Dima and Rami. I will briefly survey the method, discuss the multiplicity one result, and talk about applications to covering groups. Parts of the talk are also based on a collaboration with Cai, Friedberg and Ginzburg.
    Close details

    Foundations of Computer Science Colloquium

    Date:
    02
    Monday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Online Algorithms via Projections
    Location: Jacob Ziskind Building
    Lecturer: Seffi Naor
    Organizer: Faculty of Mathematics and Computer Science
    Details: We present a new/old approach to the design of online algorithms via Bregman pro ... Read more We present a new/old approach to the design of online algorithms via Bregman projections. This approach is applicable to a wide range of online problems and we discuss connections to previous work on online primal-dual algorithms. In particular, the k-server problem on trees and HSTs is considered. The projection-based algorithm for this problem turns out to have a competitive ratio that matches some of the recent results given by Bubeck et al. (STOC 2018), whose algorithm uses mirror-descent-based continuous dynamics prescribed via a differential inclusion. Joint work with Niv Buchbinder, Anupam Gupta, and Marco Molinaro.
    Close details

    Faculty Seminar

    Date:
    01
    Sunday
    December
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Gibbs measures vs. random walks in negative curvature
    Location: Jacob Ziskind Building
    Lecturer: Ilya Gekhtman
    Organizer: Faculty of Mathematics and Computer Science
    Details: The ideal boundary of a negatively curved manifold naturally carries two types o ... Read more The ideal boundary of a negatively curved manifold naturally carries two types of measures. On the one hand, we have conditionals for equilibrium (Gibbs) states associated to Hoelder potentials
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    28
    Thursday
    November
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: DOUBLE SEMINAR
    Location: Jacob Ziskind Building
    Lecturer: Amir Dembo
    Organizer: Faculty of Mathematics and Computer Science
    Details: Amir Dembo (Stanford) "Dynamics for spherical spin glasses: Disorder dependent i ... Read more Amir Dembo (Stanford) "Dynamics for spherical spin glasses: Disorder dependent initial conditions" Abstract: In this talk, based on a joint work with Eliran Subag, I will explain how to rigorously derive the integro-differential equations that arise in the thermodynamic limit of the empirical correlation and response functions for Langevin dynamics in mixed spherical p-spin disordered mean-field models. I will then compare the large time asymptotic of these equations in case of a uniform (infinite-temperature) starting point, to what one obtains when starting within one of the spherical bands on which the Gibbs measure
    Close details

    Vision and Robotics Seminar

    Date:
    28
    Thursday
    November
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Blind Super-Resolution Kernel Estimation using an Internal-GAN
    Location: Jacob Ziskind Building
    Lecturer: Sefi Bell Kligler
    Organizer: Faculty of Mathematics and Computer Science
    Details: Super resolution (SR) methods typically assume that the low-resolution (LR) imag ... Read more Super resolution (SR) methods typically assume that the low-resolution (LR) image was downscaled from the unknown high-resolution (HR) image by a fixed "ideal"
    Close details

    Seminar in Geometry and Topology

    Date:
    25
    Monday
    November
    2019
    Lecture / Seminar
    Time: 09:00-11:00
    Title: Zariski cancellation problem for surfaces
    Location: Jacob Ziskind Building
    Lecturer: Mikhail Zaidenberg
    Organizer: Faculty of Mathematics and Computer Science
    Details: The Zariski Cancellation Problem asks when a stable isomorphism of affine variet ... Read more The Zariski Cancellation Problem asks when a stable isomorphism of affine varieties over an algebraically closed field implies an isomorphism. This is true for affine curves (Abhyankar, Eakin, and Heinzer 72), for the affine plane in zero characteristic (Miyanishi-Sugie and Fujita 79 −80), but false for general affine surfaces in zero characteristic (Danielewski 88) and for the affine space A^3 in positive characteristic (N. Gupta 13). The talk is devoted to a recent progress in the surface case over a field of zero characteristic (Bandman-Makar-Limanov, Dubouloz, Flenner and Kaliman, e.a.). It occurs to be possible to describe the moduli space of pairs of surfaces with isomorphic cylinders.
    Close details

    Faculty Seminar

    Date:
    24
    Sunday
    November
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: On Benjamini-Schramm convergence
    Location: Jacob Ziskind Building
    Lecturer: Arie Levit
    Organizer: Faculty of Mathematics and Computer Science
    Details: Benjamini-Schramm convergence is a probabilistic notion useful in studying the a ... Read more Benjamini-Schramm convergence is a probabilistic notion useful in studying the asymptotic behavior of sequences of metric spaces. The goal of this talk is to discuss this notion and some of its applications from various perspectives, e.g. for groups, graphs, hyperbolic manifolds and locally symmetric spaces, emphasizing the distinction between the hyperbolic rank-one case and the rigid high-rank case. Understanding the "sofic" part of the Benjamini-Schramm space, i.e. all limit points of "finitary" objects, will play an important role. From the group-theoretic perspective, I will talk about sofic groups, i.e. groups which admit a probabilistic finitary approximation, as well as a companion notion of permutation stability. Several results and open problems will also be discussed.
    Close details

    Vision and Robotics Seminar

    Date:
    21
    Thursday
    November
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Tools for visual expression and communication.
    Location: Jacob Ziskind Building
    Lecturer: Ohad Fried
    Organizer: Faculty of Mathematics and Computer Science
    Details: Photos and videos are now a main mode of communication, used to tell stories, sh ... Read more Photos and videos are now a main mode of communication, used to tell stories, share experiences and convey ideas. However, common media editing tools are often either too complex to master, or oversimplified and limited.
    Close details
    Abstract: TBA ... Read more TBA
    Close abstract

    Machine Learning and Statistics Seminar

    Date:
    20
    Wednesday
    November
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Optimization's Hidden Gift to Learning: Implicit Regularization
    Location: Jacob Ziskind Building
    Lecturer: Nathan Srebro
    Organizer: Faculty of Mathematics and Computer Science
    Details: It is becoming increasingly clear that

    Faculty Seminar

    Date:
    20
    Wednesday
    November
    2019
    Lecture / Seminar
    Time: 10:00-12:00
    Title: Exploiting Tasks, Structures, and AI in Digital Communications Receivers
    Location: Jacob Ziskind Building
    Lecturer: Nir Shlezinger
    Organizer: Faculty of Mathematics and Computer Science
    Details: The broad range of demands which next generation communications systems are requ ... Read more The broad range of demands which next generation communications systems are required to meet, spanning from increased rates to efficient power consumption, cannot be satisfied by conventional architectures. These requirements give rise to a multitude of new challenges in modern communications and signal processing systems. In this talk we focus on two fundamental aspects of digital communications receivers - signal acquisition and symbol detection - discussing methods to improve their performance and reliability. For signal acquisition, we show how analog-to-digital convertors can be designed in light of the overall system task, achieving optimal-approaching performance while utilizing efficient bit-constrained samplers. Then, we discuss how recent advances in machine learning can be exploited for symbol detection, a task which is typically addressed using channel-model-based methods, such as the Viterbi algorithm and interference cancellation methods. We present a new framework for combining artificial intelligence and model-based algorithms, allowing the latter to be implemented in a data-driven manner. The resulting architectures are capable of approaching the established performance of model-based algorithms, while being invariant of the underlying statistical model, and learning it from a small amount of training samples.
    Close details

    Seminar in Geometry and Topology

    Date:
    19
    Tuesday
    November
    2019
    Lecture / Seminar
    Time: 16:00-18:00
    Title: Topological recursion for Masur-Veech volumes
    Location: Jacob Ziskind Building
    Lecturer: Gaetan Borot's
    Organizer: Faculty of Mathematics and Computer Science
    Details: The moduli space of quadratic differentials on punctured Riemann surfaces admits ... Read more The moduli space of quadratic differentials on punctured Riemann surfaces admits a integral piecewise linear structure. It gives rise to a measure whose total mass is finite and called the Masur-Veech volume. These volumes are also related to the asymptotic growth of the number of multicurves on hyperbolic surfaces, once averaged against the Weil-Petersson metric. Adopting this point of view, we show that the Masur-Veech volumes can be obtained as the constant term in a family of polynomials computed by the topological recursion, and as integration of a class on the moduli space of curves. Based on work of Moeller, we can prove the same statement but for a different family of polynomial (and different initial data for the topological recursion). This is based on joint works with Jorgen Andersen, Severin Charbonnier, Vincent Delecroix, Alessandro Giacchetto, Danilo Lewanski and Campbell Wheeler.
    Close details

    Special Guest Lecture

    Date:
    19
    Tuesday
    November
    2019
    Lecture / Seminar
    Time: 11:15-12:05
    Title: Smooth models of systems with complicated stochastic behaviors
    Location: Jacob Ziskind Building
    Lecturer: Professor Yakov Pesin
    Organizer: Faculty of Mathematics and Computer Science
    Details: To describe different levels of stochastic behavior Ergodic Theory introduces a ... Read more To describe different levels of stochastic behavior Ergodic Theory introduces a hierarchy of ergodic properties. Among them ergodicity, mixing and the Bernoulli property are most used to study models in science with complicated "turbulent-like" behavior. However, "typical" systems in science are modeled by smooth (differentiable) dynamical systems acting on smooth phase spaces and the classical "smooth realization problem" asks whether any smooth phase space allows a dynamical system with prescribe collection of ergodic (stochastic) properties, in other words whether topology of the phase space may impose obstructions on a system to exhibit certain stochastic behavior. In this connection I also discuss two more important characteristics of stochastic behavior -- the rate of decay of correlations (rate of mixing) and the Central Limit Theorem -- and whether they too can be realized on any smooth phase space.
    Close details

    Foundations of Computer Science Colloquium

    Date:
    18
    Monday
    November
    2019
    Lecture / Seminar
    Time: 11:15-12:45
    Title: Privately Learning Thresholds: Closing the Exponential Gap
    Location: Jacob Ziskind Building
    Lecturer: Uri Stemmer
    Organizer: Faculty of Mathematics and Computer Science
    Details: We study the sample complexity of learning threshold functions under the constra ... Read more We study the sample complexity of learning threshold functions under the constraint of differential privacy. Unlike the non-private case, where the sample complexity is independent of the domain size, it turns our that for private learning the sample complexity must depend on the domain size $X$. Our current understanding of this task places its sample complexity somewhere between $log^*|X|$ and $2^{log^*|X|}$, where at least three different algorithms are known with sample complexity exponential in $log^*|X|$. In this work we reduce this gap significantly, and show that the sample complexity is at most polynomial in $log^*|X|$. Joint work with Haim Kaplan, Katrina Ligett, Yishay Mansour, and Moni Naor.
    Close details

    Seminar in Geometry and Topology

    Date:
    18
    Monday
    November
    2019
    Lecture / Seminar
    Time: 09:00-11:00
    Title: Open Gromov-Witten invariants and underlying structures
    Location: Jacob Ziskind Building
    Lecturer: Sara Tukachinsky
    Organizer: Faculty of Mathematics and Computer Science
    Details: For X a symplectic manifold and L a Lagrangian submanifold, genus zero open Grom ... Read more For X a symplectic manifold and L a Lagrangian submanifold, genus zero open Gromov-Witten (OGW) invariants count configurations of pseudoholomorphic disks in X with boundary conditions in L and various constraints at boundary and interior marked points. In a joint work with Jake Solomon from 2016, we define OGW invariants using bounding chains, a concept that comes from Floer theory. In a recent work, also joint with Solomon, we find that the generating function of OGW satisfies a system of PDE called open WDVV equation. This PDE translates to an associativity relation for a quantum product we define on the relative cohomology H^*(X,L). For the projective space, open WDVV gives rise to recursions that, together with other properties, allow the computation of all OGW invariants.
    Close details

    Faculty Seminar

    Date:
    17
    Sunday
    November
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: Automated Design of Better Environments for Intelligent Agents
    Location: Jacob Ziskind Building
    Lecturer: Sarah Keren
    Organizer: Faculty of Mathematics and Computer Science
    Details: Automated agents and humans increasingly interact and collaborate: at home, at t ... Read more Automated agents and humans increasingly interact and collaborate: at home, at the workplace, on the road, and in many other everyday settings. In all these settings, effectively recognizing what users try to achieve, providing relevant assistance (or, depending on the application, taking relevant preventive measures), and supporting an effective collaboration in the system are essential tasks. All these tasks can be enhanced via efficient system redesign, and often even subtle changes can yield great benefits. However, since these systems are typically complex, hand crafting good design solutions is hard. Utility Maximizing Design (UMD) addresses this challenge by automating the design process. It does so by offering informed search strategies to automatically and efficiently find optimal design solutions for maximizing a variety of targeted objectives. One example is Goal Recognition Design (GRD), which seeks a redesign to an environment that minimizes the maximal progress an agent can make before its goal is revealed. A second is
    Close details

    Faculty Seminar

    Date:
    17
    Sunday
    November
    2019
    Lecture / Seminar
    Time: 11:15-13:00
    Title: TBA
    Location: Jacob Ziskind Building
    Lecturer: Sarah Keren
    Organizer: Faculty of Mathematics and Computer Science
    Details: TBA

    Geometric Functional Analysis and Probability Seminar

    Date:
    14
    Thursday
    November
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Random Cayley graphs
    Location: Jacob Ziskind Building
    Lecturer: Jonathan Hermon
    Organizer: Faculty of Mathematics and Computer Science
    Details: We consider the random Cayley graph of a finite group $G$ formed by picking $k$ ... Read more We consider the random Cayley graph of a finite group $G$ formed by picking $k$ random generators uniformly at random: We prove universality of cutoff (for the random walk) and a concentration of measure phenomenon in the Abelian setup (namely, that all but $o(|G|)$ elements lie at distance $[R-o(R),R-o(R)]$ from the origin, where $R$ is the minimal ball in $Z^k$ of size at least $|G|$), provided $k gg 1$ is large in terms of the size of the smallest generating set of $G$. As conjectured by Aldous and Diaconis, the cutoff time is independent of the algebraic structure (it is given by the time at which the entropy of a random walk on $Z^k$ is $log|G|$). We prove analogous results for the Heisenberg $H_{p,d}$
    Close details

    Seminar in Geometry and Topology

    Date:
    12
    Tuesday
    November
    2019
    Lecture / Seminar
    Time: 16:00-18:00
    Title: Open Gromov-Witten invariants and underlying structures
    Location: Jacob Ziskind Building
    Lecturer: Sara Tukachinsky
    Organizer: Faculty of Mathematics and Computer Science
    Details: For X a symplectic manifold and L a Lagrangian submanifold, genus zero open Grom ... Read more For X a symplectic manifold and L a Lagrangian submanifold, genus zero open Gromov-Witten (OGW) invariants count configurations of pseudoholomorphic disks in X with boundary conditions in L and various constraints at boundary and interior marked points. In a joint work with Jake Solomon from 2016, we define OGW invariants using bounding chains, a concept that comes from Floer theory. In a recent work, also joint with Solomon, we find that the generating function of OGW satisfies a system of PDE called open WDVV equation. This PDE translates to an associativity relation for a quantum product we define on the relative cohomology H^*(X,L). For the projective space, open WDVV gives rise to recursions that, together with other properties, allow the computation of all OGW invariants.
    Close details

    Mathematical Analysis and Applications Seminar

    Date:
    12
    Tuesday
    November
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Revisiting Totally Positive Differential Systems: A Tutorial and New Results
    Location: Jacob Ziskind Building
    Lecturer: Michael Margaliot
    Organizer: Faculty of Mathematics and Computer Science
    Details: A matrix is called totally nonnegative (TN) if all its minors are

    Geometric Functional Analysis and Probability Seminar

    Date:
    31
    Thursday
    October
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Geometry of transportation cost (a.k.a. Earth Mover or Wasserstein distance)
    Location: Jacob Ziskind Building
    Lecturer: Mikhail Ostrovskii
    Organizer: Faculty of Mathematics and Computer Science
    Details: We consider (finitely supported) transportation problems on a metric space M. Th ... Read more We consider (finitely supported) transportation problems on a metric space M. They form a vector space TP(M). The optimal transportation cost for such transportation problems is a norm on this space. This normed space is of interest for the theory of metric embeddings because the space M embeds into it isometrically. I am going to talk about geometry of such normed spaces. The most important questions for this talk are relations of these spaces with $L_1$ and $L_infty$ spaces.
    Close details

    Seminar in Geometry and Topology

    Date:
    28
    Monday
    October
    2019
    Lecture / Seminar
    Time: 09:00-11:00
    Title: Cantor-Bendixson rank of meromorphic differentials
    Location: Jacob Ziskind Building
    Lecturer: Guillaume Tahar
    Organizer: Faculty of Mathematics and Computer Science
    Details: Meromorphic differentials define flat metrics with singularities on Riemann surf ... Read more Meromorphic differentials define flat metrics with singularities on Riemann surfaces. Directions of geodesic segments between singularities are a closed subset of the circle. The Cantor-Bendixson rank of their set of directions is a descriptive set-theoretic measure complexity. Drawing on a previous work of David Aulicino, we prove a sharp upper bound that depends only on the genus of the underlying topological surface.
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    24
    Thursday
    October
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Transport bounds for empirical measures
    Location: Jacob Ziskind Building
    Lecturer: Sergey G. Bobkov
    Organizer: Faculty of Mathematics and Computer Science
    Details: We will be discussing a Fourier-analytic approach to optimal matching between in ... Read more We will be discussing a Fourier-analytic approach to optimal matching between independent samples, with an elementary proof of the Ajtai-Komlos-Tusnady theorem. The talk is based on a joint work with Michel Ledoux.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    24
    Tuesday
    September
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Noncommutative Catalan numbers
    Location: Jacob Ziskind Building
    Lecturer: Arkady Berenstein
    Organizer: Faculty of Mathematics and Computer Science
    Details: The goal of my talk (based on joint work with Vladimir Retakh) is to introduce n ... Read more The goal of my talk (based on joint work with Vladimir Retakh) is to introduce noncommutative analogs of Catalan numbers c_n which belong to the free Laurent polynomial algebra L_n in n generators. Our noncommutative Catalan numbers C_n admit interesting (commutative and noncommutative) specializations, one of them related to Garsia-Haiman (q,t)-versions, another -- to solving noncommutative quadratic equations. We also establish total positivity of the corresponding (noncommutative) Hankel matrices H_n and introduce two kinds of noncommutative binomial coefficients which are instrumental in computing the inverse of H_n and its positive factorizations, and other combinatorial identities involving C_n. If time permits, I will explain the relationship of the C_n with the: 1. noncommutative Laurent Phenomenon, which was previously established for Kontsevich rank 2 recursions and all marked surfaces 2. noncommutative orthogonal polynomials, which can be viewed as noncommutative determinants of an extended matrix H_n.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    11
    Wednesday
    September
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: How zero is zero?
    Location: Jacob Ziskind Building
    Lecturer: Thorsten Heidersdorf
    Organizer: Faculty of Mathematics and Computer Science
    Details: The quotient of a monoidal category by its largest tensor ideal - given by the s ... Read more The quotient of a monoidal category by its largest tensor ideal - given by the so-called negligible morphisms - is often a semisimple category. I will introduce a generalization of the notion of negligible morphism for some monoidal categories and discuss the associated tensor ideals in the setting of Deligne categories and tilting modules for quantum groups and algebraic groups. It turns out that they are related to other notions from representation theory like modified dimensions and the a-function.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    10
    Tuesday
    September
    2019
    Lecture / Seminar
    Time: 10:15-12:30
    Title: Representations of Coulomb branches and Gelfand-Tsetlin modules
    Location: Jacob Ziskind Building
    Lecturer: Oded Yacobi
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let g be a Lie algebra of type ADE. To a pair of weights of g (one dominant, the ... Read more Let g be a Lie algebra of type ADE. To a pair of weights of g (one dominant, the other arbitrary) we associate a group G and a representation N consisting of framed quiver representations of the Dynkin diagram of g. From (G,N) we can construct two varieties. The Higgs branch is the categorical quotient of N by G, which in this case is the Nakajima quiver variety and has been studied for over 25 years. The Coulomb branch has a much more complicated definition that was only recently discovered by Braverman, Finkelberg, and Nakajima. There is a duality between these spaces, which is sometimes referred to as 3d mirror symmetry or symplectic duality. In this talk I'll try to explain the definition of the Coulomb branch, and why you might care. I will discuss its deformation quantization, which appears naturally from the construction. I'll describe also our recent result which provides an equivalence between representations of the deformation quantisation, and modules over a seemingly very different algebra which is defined combinatorially and arises in categorical representation theory. This equivalence has several interesting consequences, e.g. it provides a classification for the irreducible Gelfand-Tsetlin modules of gl(n), which was previously only known up to n=3.
    Close details

    Vision and Robotics Seminar

    Date:
    08
    Sunday
    September
    2019
    Lecture / Seminar
    Time: 11:15-13:15
    Title: Adapting and Explaining Deep Learning for Autonomous Systems
    Location: Jacob Ziskind Building
    Lecturer: Trevor Darrell
    Organizer: Faculty of Mathematics and Computer Science
    Details: Learning of layered or "deep" representations has recently enabled low-cost sens ... Read more Learning of layered or "deep" representations has recently enabled low-cost sensors for autonomous vehicles and efficient automated analysis of visual semantics in online media. But these models have typically required prohibitive amounts of training data, and thus may only work well in the environment they have been trained in.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    20
    Tuesday
    August
    2019
    Lecture / Seminar
    Time: 11:15-12:15
    Title: Classifying certain classes of braid representations
    Location: Jacob Ziskind Building
    Lecturer: Hans Wenzl
    Organizer: Faculty of Mathematics and Computer Science
    Details: Representations of braid groups have appeared in many different areas such as to ... Read more Representations of braid groups have appeared in many different areas such as topology, statistical mechanics, conformal field theory, braided tensor categories and others. In order to compare these, intrinsic characterizations of such representations are desirable. These have been known for some time for representations in connection with vector representations of classical Lie types, in terms of Hecke algebras and so-called BMW algebras. We review these and show how these results can be extended to include more cases related to exceptional Lie types. In particular, we obtain new classes of braid representations where the images of the generators satisfy a cubic equation. Time permitting, we discuss applications of these results such as Schur-Weyl type duality theorems and classification of braided tensor categories.
    Close details

    Machine Learning and Statistics Seminar

    Date:
    13
    Tuesday
    August
    2019
    Lecture / Seminar
    Time: 14:15-16:00
    Title: The Information in the Weights of a Deep Network, and its consequences for transfer learning and critical learning periods
    Location: Jacob Ziskind Building
    Lecturer: Stefano Soatto
    Organizer: Faculty of Mathematics and Computer Science
    Abstract: The information in the weights of deep networks plays a key role in understandin ... Read more The information in the weights of deep networks plays a key role in understanding their behavior. When used as a regularizer during training (either explicitly or implicitly) it induces generalization through the PAC-Bayes bound. Rather than being computed and minimized explicitly, it can be directly controlled by injecting noise in the weights, a process known as Information Dropout. It can also be shown that stochastic gradient descent (SGD), when taken to the continuous limit and interpreted in the Wasserstein metric, adds the information in the weights as inductive bias, even if not explicitly present in the loss function. The Emergence Bound shows that, provided that a trained network has sufficient capacity, minimizing the information in the weights, which is a function of the training set consisting of data seen in the past, guarantees minimality, invariance, and independence of the components of the representation of the test (future) data. The trace of the information in the weights during training shows that, rather than increasingly monotonically through the learning process, as one would expect, first increases rapidly in the first few epochs, and then decreases to an asymptote that is a fraction of its peak. This unusual behavior qualitatively follows the sensitivity to critical learning periods observed in biological systems, from cats to humans, as well as recently in deep artificial networks. Together with the Emergence Bound and the PAC-Bayes bound, this shows that forgetting is a necessary part of the learning process, and happens while precision increases monotonically. The information in the weights can also be used to defined a topology in the space of tasks, and an asymmetric distance that can be used to predict the cost and performance of fine-tuning a network trained for a different task, without actually performing the experiment. These phenomena collectively point to the importance of the dynamics of learning, and suggests that studying the transient behavior can yield insight beyond those that can be gleaned from the asymptotics. Depending on the context, we use Shannon, Fisher, or Kolmogorov information to prove the results described.
    Close abstract

    Vision and Robotics Seminar

    Date:
    04
    Sunday
    August
    2019
    Lecture / Seminar
    Time: 13:15-14:30
    Title: On the Implicit Bias of Dropout
    Location: Jacob Ziskind Building
    Lecturer: Rene Vidal
    Organizer: Faculty of Mathematics and Computer Science
    Details: Dropout is a simple yet effective regularization technique that has been applied ... Read more Dropout is a simple yet effective regularization technique that has been applied to various machine learning tasks
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    25
    Thursday
    July
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Anchored expansion in supercritical percolation on nonamenable graphs.
    Location: Jacob Ziskind Building
    Lecturer: Jonathan Hermon
    Organizer: Faculty of Mathematics and Computer Science
    Details: Let G be a transitive nonamenable graph, and consider supercritical Bernoulli bo ... Read more Let G be a transitive nonamenable graph, and consider supercritical Bernoulli bond percolation on G. We prove that the probability that the origin lies in a finite cluster of size n decays exponentially in n. We deduce that: Every infinite cluster has anchored expansion almost surely. This answers positively a question of Benjamini, Lyons, and Schramm (1997). Various observables, including the percolation probability and the truncated susceptibility are analytic functions of p throughout the entire supercritical phase. Joint work with Tom Hutchcroft.
    Close details

    Algebraic Geometry and Representation Theory Seminar

    Date:
    16
    Tuesday
    July
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Cacti and spectra of monomial braidings
    Location: Jacob Ziskind Building
    Lecturer: Jacob Greenstein
    Organizer: Faculty of Mathematics and Computer Science

    Geometric Functional Analysis and Probability Seminar

    Date:
    11
    Thursday
    July
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Harmonic Analysis on GL_n over finite fields
    Location: Jacob Ziskind Building
    Lecturer: Shamgar Gurevitch
    Organizer: Faculty of Mathematics and Computer Science
    Details: There are many formulas that express interesting properties of a finite group G ... Read more There are many formulas that express interesting properties of a finite group G in terms of sums over its characters. For evaluating or estimating these sums, one of the most salient quantities to understand is the {it character ratio}: $$ trace( ho(g))/dim( ho), $$ for an irreducible representation $ ho$ of G and an element g of G. For example, Diaconis and Shahshahani stated a formula of the mentioned type for analyzing certain random walks on G. Recently, we discovered that for classical groups G over finite fields there is a natural invariant of representations that provides strong information on the character ratio. We call this invariant rank. This talk will discuss the notion of rank for GLn over finite fields, and explain how one can apply the results to verify mixing time and rate for certain random walks. The talk will assume basic notions of linear algebra in Hilbert spaces, and the definition of a group. This is joint work with Roger Howe (Yale and Texas AM).
    Close details

    Vision and Robotics Seminar

    Date:
    11
    Thursday
    July
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: A Monte Carlo Framework for Rendering Speckle Statistics in Scattering Media
    Location: Jacob Ziskind Building
    Lecturer: Anat Levin
    Organizer: Faculty of Mathematics and Computer Science
    Details: We present a Monte Carlo rendering framework for the physically-accurate simulat ... Read more We present a Monte Carlo rendering framework for the physically-accurate simulation of speckle patterns arising from volumetric scattering of coherent waves. These noise-like patterns are characterized by strong statistical properties, such as the so-called memory effect. These properties are at the core of imaging techniques for applications as diverse as tissue imaging, motion tracking, and non-line-of-sight imaging. Our rendering framework can replicate these properties computationally, in a way that is orders of magnitude more efficient than alternatives based on directly solving the wave equations. At the core of our framework is a path-space formulation for the covariance of speckle patterns arising from a scattering volume, which we derive from first principles. We use this formulation to develop two Monte Carlo rendering algorithms, for computing speckle covariance as well as directly speckle fields. While approaches based on wave equation solvers require knowing the microscopic position of wavelength-sized scatterers, our approach takes as input only bulk parameters describing the statistical distribution of these scatterers inside a volume. We validate the accuracy of our framework by comparing against speckle patterns simulated using wave equation solvers, use it to simulate memory effect observations that were previously only possible through lab measurements, and demonstrate its applicability for computational imaging tasks.
    Close details
    Abstract: ... Read more
    Close abstract

    Foundations of Computer Science Seminar

    Date:
    08
    Monday
    July
    2019
    Lecture / Seminar
    Time: 14:30-16:00
    Title: The Mechanism Design Approach to Interactive Proofs
    Location: Jacob Ziskind Building
    Lecturer: Shikha Singh
    Organizer: Faculty of Mathematics and Computer Science
    Details: The model of interactive proofs, introduced nearly two and a half decade, is now ... Read more The model of interactive proofs, introduced nearly two and a half decade, is now increasingly widely being used to design computation-outsourcing protocols. In an interactive proof, an honest party interacts with powerful but strategic provers, to elicit from them the correct answer to a computational question. Classical interactive proofs assume that the provers are adversarial (i.e., they want to mislead the verifier) and cooperative (work together as a team). In this talk, I will present my work on a new payment-based interactive-proof system, called rational proofs. In rational proofs, the provers are not adversarial but rational, that is, they want to maximize the payment received from the verifier. Using principles from mechanism design, I will show how these payments can be used to leverage correctness from multiple provers who are either cooperative or non-cooperative in nature. I will also present how the guarantees of rational proofs are related to the soundness and completeness guarantees of classical interactive proofs. Bio: Shikha Singh is currently an Assistant Professor of Computer Science at Wellesley College and will be joining Williams College as an Assistant Professor in Fall 2019. She obtained her PhD in Computer Science from Stony Brook University and her Integrated MSc. in Mathematics and Computing from IIT Kharagpur. Her broad research interests include algorithmic game theory, algorithms and data structures for big data, and complexity theory.
    Close details
    Abstract: The model of interactive proofs, introduced nearly two and a half decade, is now ... Read more The model of interactive proofs, introduced nearly two and a half decade, is now increasingly widely being used to design computation-outsourcing protocols. In an interactive proof, an honest party interacts with powerful but strategic provers, to elicit from them the correct answer to a computational question. Classical interactive proofs assume that the provers are adversarial (i.e., they want to mislead the verifier) and cooperative (work together as a team). In this talk, I will present my work on a new payment-based interactive-proof system, called rational proofs. In rational proofs, the provers are not adversarial but rational, that is, they want to maximize the payment received from the verifier. Using principles from mechanism design, I will show how these payments can be used to leverage correctness from multiple provers who are either cooperative or non-cooperative in nature. I will also present how the guarantees of rational proofs are related to the soundness and completeness guarantees of classical interactive proofs. Bio: Shikha Singh is currently an Assistant Professor of Computer Science at Wellesley College and will be joining Williams College as an Assistant Professor in Fall 2019. She obtained her PhD in Computer Science from Stony Brook University and her Integrated MSc. in Mathematics and Computing from IIT Kharagpur. Her broad research interests include algorithmic game theory, algorithms and data structures for big data, and complexity theory.
    Close abstract

    Geometric Functional Analysis and Probability Seminar

    Date:
    04
    Thursday
    July
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Stationary Hastings-Levitov model
    Location: Jacob Ziskind Building
    Lecturer: Eviatar Procaccia
    Organizer: Faculty of Mathematics and Computer Science
    Details: We construct and study a stationary version of the Hastings-Levitov(0) model. We ... Read more We construct and study a stationary version of the Hastings-Levitov(0) model. We prove that unlike the classical model, in the stationary case particle sizes are constant in expectation, yielding that this model can be seen as a tractable off-lattice Diffusion Limited Aggregation (DLA). The stationary setting together with a geometric interpretations of the harmonic measure yields new geometric results such as stabilization, finiteness of arms and unbounded width in mean of arms. Moreover we can present an exact conjecture for the fractal dimension.
    Close details

    Vision and Robotics Seminar

    Date:
    04
    Thursday
    July
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Linearly Converging Quasi Branch and Bound Algorithms for Global Rigid Registration
    Location: Jacob Ziskind Building
    Lecturer: Nadav Dym
    Organizer: Faculty of Mathematics and Computer Science
    Details:
    Abstract: Rigid registration is  the problem of finding the optimal rigid motion and corr ... Read more Rigid registration is  the problem of finding the optimal rigid motion and correspondence between two shapes, so that they are as similar as possible in terms of an appropriate energy. We will describe several popular algorithms for this problem: PCA alignment and ICP, which are very efficient but are not globally optimal, as well as sampling and branch and bound (BnB) algorithms which exhibit slow convergence but are globally optimal. Next we suggest our quasi BnB algorithm as an improvement upon the BnB approach. Quasi BnB replaces the linear lower bounds used in BnB algorithms with quadratic quasi-lower bounds. While quasi-lower bounds are not truly lower bounds, the Quasi-BnB algorithm is globally optimal. Our experiments show that Quasi-BnB is dramatically more efficient than BnB algorithms. Theoretically we show that quasi-BnB exhibits linear convergence -- it achieves ϵ-accuracy in O(log(1/ϵ)) time while the time complexity of other rigid registration BnB algorithms is polynomial in 1/ϵ.
    Close abstract

    Machine Learning and Statistics Seminar

    Date:
    03
    Wednesday
    July
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Learning Linear-Quadratic Regulators Efficiently with only $sqrt{T}$ Regret
    Location: Jacob Ziskind Building
    Lecturer: Alon Cohen
    Organizer: Faculty of Mathematics and Computer Science
    Details: The linear-quadratic regulator is a simple and ubiquitous model in optimal contr ... Read more The linear-quadratic regulator is a simple and ubiquitous model in optimal control. Classical results in control theory pertain to asymptotic convergence and stability of such systems. Recently, it has received renewed interest from a learning-theoretic perspective with a focus on computational tractability and finite-time convergence guarantees. Among the most challenging problems in LQ control is that of adaptive control: regulating a system with parameters that are initially unknown and have to be learned while incurring the associated costs. In its modern incarnation, this problem is approached through the lens of online learning with partial feedback (e.g., multi-armed bandits) and regret minimization. Still, recent results derive algorithms that are either computationally intractable or suffer from high regret. In this talk, I will present the first computationally-efficient algorithm with $widetilde O(sqrt{T})$ regret for learning in linear quadratic control systems with unknown dynamics. By that, this resolves an open question of Abbasi-Yadkori and Szepesvari (2011) and Dean, Mania, Matni, Recht, and Tu (2018). The key to the efficiency of our algorithm is in a novel reformulation of the LQ control problem as a convex semi-definite program. This is joint work with Tomer Koren and Yishay Mansour presented at ICML 2019.
    Close details

    Vision and Robotics Seminar

    Date:
    27
    Thursday
    June
    2019
    Lecture / Seminar
    Time: 12:15-13:30
    Title: Exploring the Bounds of the Utility of Context for Object Detection
    Location: Jacob Ziskind Building
    Lecturer: Ehud Barnea
    Organizer: Faculty of Mathematics and Computer Science
    Details: The recurring context in which objects appear holds valuable information that ca ... Read more The recurring context in which objects appear holds valuable information that can be employed to predict their existence. This intuitive observation indeed led many researchers to endow appearance-based detectors with explicit reasoning about context. The underlying thesis suggests that stronger contextual relations would facilitate greater improvements in detection capacity. In practice, however, the observed improvement in many cases is modest at best, and often only marginal. In this work we seek to improve our understanding of this phenomenon, in part by pursuing an opposite approach. Instead of attempting to improve detection scores by employing context, we treat the utility of context as an optimization problem: to what extent can detection scores be improved by considering context or any other kind of additional information? With this approach we explore the bounds on improvement by using contextual relations between objects and provide a tool for identifying the most helpful ones. We show that simple co-occurrence relations can often provide large gains, while in other cases a significant improvement is simply impossible or impractical with either co-occurrence or more precise spatial relations. To better understand these results we then analyze the ability of context to handle different types of false detections, revealing that tested contextual information cannot ameliorate localization errors, severely limiting its gains. These and additional insights further our understanding on where and why utilization of context for object detection succeeds and fails
    Close details

    Geometric Functional Analysis and Probability Seminar

    Date:
    20
    Thursday
    June
    2019
    Lecture / Seminar
    Time: 13:30-15:30
    Title: Double seminar
    Location: Jacob Ziskind Building
    Lecturer: Mark Rudelson (UMich)
    Organizer: Faculty of Mathematics and Computer Science
    Details: Speaker 1: Mark Rudelson (UMich) Title: Circular law for sparse random matric ... Read more Speaker 1: Mark Rudelson (UMich) Title: Circular law for sparse random matrices. Abstract: Consider a sequence of $n$ by $n$ random matrices $A_n$ whose entries are independent identically distributed random variables. The circular law asserts that the distribution of the eigenvalues of properly normalized matrices $A_n$ converges to the uniform measure on the unit disc as $n$ tends to infinity. We prove this law for sparse random matrices under the optimal sparsity assumption. Joint work with Konstantin Tikhomirov. Speaker 2: Serguei Popov (IMECC) Title: On the range of a two-dimensional conditioned random walk Abstract: We consider the two-dimensional simple random walk conditioned on never hitting the origin. This process is a Markov chain, namely, it is the Doob $h$-transform of the simple random walk with respect to the potential kernel. It is known to be transient and we show that it is "almost recurrent'' in the sense that each infinite set is visited infinitely often, almost surely. We prove that, for a "typical large set", the proportion of its sites visited by the conditioned walk is approximately a Uniform$[0,1]$ random variable. Also, given a set $GsubsetR^2$ that does not "surround" the origin, we prove that a.s. there is an infinite number of $k$'s such that $kGcap ^2$ is unvisited. These results suggest that the range of the conditioned walk has "fractal" behavior. This is a joint work with Nina Gantert and Marina Vachkovskaia, see arxiv.org/abs/1804.00291 Also, there is much more about conditioned walks in my new book (www.ime.unicamp.br/~popov/2srw.pdf, work in progress). Comments and suggestions on the latter are very welcome!
    Close details

    Machine Learning and Statistics Seminar

    Date:
    19
    Wednesday
    June
    2019
    Lecture / Seminar
    Time: 11:15-12:30
    Title: Theoretical and Empirical Investigation of Several Common Practices in Deep Learning
    Location: Jacob Ziskind Building
    Lecturer: Daniel Soudry
    Organizer: Faculty of Mathematics and Computer Science
    Details: We examine several empirical and theoretical results on the training of deep net ... Read more We examine several empirical and theoretical results on the training of deep networks. For example, Why are common "over-fitting" indicators (e.g., very low training error, high validation loss) misleading? Why, sometimes, early-stopping time never arrives? Why can adaptive rate methods (e.g., adam) degrade generalization? Why commonly used loss functions exhibit better generalization than others? Why use weight decay before batch-norm? When can we use low numerical precision, and how low can we get? and discuss the practical implications of these results. Bio == Since October 2017, Daniel soudry is an assistant professor (Taub Fellow) in the Department of Electrical Engineering at the Technion, working in the areas of machine learning and theoretical neuroscience. Before that, he did his post-doc (as a Gruss Lipper fellow) working with Prof. Liam Paninski in the Department of Statistics, the Center for Theoretical Neuroscience the Grossman Center for Statistics of the Mind at Columbia University. He did his Ph.D. in the Department of Electrical Engineering at the Technion, Israel Institute of technology, under the guidance of Prof. Ron Meir. He received his B.Sc. degree in Electrical Engineering and Physics from the Technion.
    Close details