|
Speaker: Enno Keßler, Max Planck Institute for Mathematics (Bonn)Title: An operadic structure on supermoduli spacesVenue: VirtualAbstract: The operadic structure on the moduli spaces of algebraic curves encodes in a combinatorial way how nodal curves in the boundary can be obtained by glueing smooth curves along marked points. In this talk, I will present a generalization of the operadic structure to moduli spaces of SUSY curves (or super Riemann surfaces). This requires colored graphs and generalized operads in the sense of Borisov-Manin. Based joint work with Yu. I. Manin and Y. Wu. https://arxiv.org/abs/2202.10321 |
|
Speaker: Kenji Kawaguchi, National University of SingaporeTitle: On optimization and generalization in deep learningVenue: VirtualAbstract: Deep neural networks have achieved significant empirical success in many fields, including the fields of computer vision and natural language processing. Along with its empirical success, deep learning has been theoretically shown to be attractive in terms of its expressive power. However, the theory of expressive power does not ensure that we can efficiently find an optimal solution in terms of optimization and generalization, during the optimization process. In this talk, I will discuss some mathematical properties of optimization and generalization for deep neural networks. |
|
Speaker: Wei Ai, University of Maryland, College ParkTitle: Virtual Teams in Gig Economy — An End-to-End Data Science ApproachVenue: VirtualAbstract: The gig economy provides workers with the benefits of autonomy and flexibility, but it does so at the expense of work identity and co-worker bonds. Among the many reasons why gig workers leave their platforms, an unexplored aspect is the organization identity. In a series of studies, we develop a team formation and inter-team contest at a ride-sharing platform. We employ an end-to-end data science approach, combining methodologies from randomized field experiments, recommender systems, and counterfactual machine learning. Together, our results show that platform designers can leverage team identity and team contests to increase revenue and worker engagement in a gig economy. Bio: Wei Ai is an Assistant Professor in the College of Information Studies (iSchool) and the Institute… |
|
Speaker: Rong Ge, Duke UniversityTitle: Towards Understanding Training Dynamics for Mildly Overparametrized ModelsVenue: VirtualAbstract: While over-parameterization is widely believed to be crucial for the success of optimization for the neural networks, most existing theories on over-parameterization do not fully explain the reason — they either work in the Neural Tangent Kernel regime where neurons don’t move much, or require an enormous number of neurons. In this talk I will describe our recent works towards understanding training dynamics that go beyond kernel regimes with only polynomially many neurons (mildly overparametrized). In particular, we first give a local convergence result for mildly overparametrized two-layer networks. We then analyze the global training dynamics for a related overparametrized tensor model. For both works, we rely on a key intuition that neurons in overparametrized models work in… |
|
Speaker: Hui Yu, National University of SingaporeTitle: Singular Set in Obstacle ProblemsVenue: VirtualAbstract: In this talk we describe a new method to study the singular set in the obstacle problem. This method does not depend on monotonicity formulae and works for fully nonlinear elliptic operators. The result we get matches the best-known result for the case of Laplacian. |
|
Speaker: Guilherme Ost, Institute of Mathematics of the Federal University of Rio de JaneiroTitle: Sparse Markov Models for High-dimensional InferenceVenue: VirtualAbstract: Finite order Markov models are theoretically well-studied models for dependent data. Despite their generality, application in empirical work when the order is larger than one is quite rare. Practitioners avoid using higher order Markov models because (1) the number of parameters grow exponentially with the order, (2) the interpretation is often difficult. Mixture of transition distribution models (MTD) were introduced to overcome both limitations. MTD represent higher order Markov models as a convex mixture of single step Markov chains, reducing the number of parameters and increasing the interpretability. Nevertheless, in practice, estimation of MTD models with large orders are still limited because of curse of dimensionality and high algorithm complexity. Here, we prove that if only few lags… |
|
Speaker: Maddie Weinstein, Stanford UniversityTitle: 2/10/2022 – Interdisciplinary Science SeminarVenue: VirtualTitle: Metric Algebraic Geometry Abstract: A real algebraic variety is the set of points in real Euclidean space that satisfy a system of polynomial equations. Metric algebraic geometry is the study of properties of real algebraic varieties that depend on a distance metric. In this talk, we introduce metric algebraic geometry through a discussion of Voronoi cells, bottlenecks, and the reach of an algebraic variety. We also show applications to the computational study of the geometry of data with nonlinear models. |
|
Speaker: Aaron Fenyes, Institut des Hautes Études ScientifiquesTitle: 2/3/2022 – Interdisciplinary Science SeminarVenue: VirtualTitle:Quasiperiodic prints from triply periodic blocks Abstract: Slice a triply periodic wooden sculpture along an irrational plane. If you ink the cut surface and press it against a page, the pattern you print will be quasiperiodic. Patterns like these help physicists see how metals conduct electricity in strong magnetic fields. I’ll show you some block prints that imitate the printing process described above, and I’ll point out the visual features that reveal conductivity properties. Interactive slides:https://www.ihes.fr/~fenyes/seeing/slices/ |
|
Speaker: Fabian Gundlach, Harvard UniversityTitle: 1/27/2022 – Interdisciplinary Science SeminarVenue: VirtualTitle: Polynomials vanishing at lattice points in convex sets Abstract: Let P be a convex subset of R^2. For large d, what is the smallest degree r_d of a polynomial vanishing at all lattice points in the dilate d*P? We show that r_d / d converges to some positive number, which we compute for many (but maybe not all) triangles P. |
|
Speaker:Title: 1/20/2022 – Interdisciplinary Science SeminarVenue: VirtualTitle: Markov chains, optimal control, and reinforcement learning Abstract: Markov decision processes are a model for several artificial intelligence problems, such as games (chess, Go…) or robotics. At each timestep, an agent has to choose an action, then receives a reward, and then the agent’s environment changes (deterministically or stochastically) in response to the agent’s action. The agent’s goal is to adjust its actions to maximize its total reward. In principle, the optimal behavior can be obtained by dynamic programming or optimal control techniques, although practice is another story. Here we consider a more complex problem: learn all optimal behaviors for all possible reward functions in a given environment. Ideally, such a “controllable agent” could be given a description… |
|
Speaker: Boyu Zhang, Princeton UniversityTitle: 1/6/2022 Interdisciplinary Science SeminarVenue: VirtualTitle: The smooth closing lemma for area-preserving surface diffeomorphisms Abstract: In this talk, I will introduce the smooth closing lemma for area-preserving diffeomorphisms on surfaces. The proof is based on a Weyl formula for PFH spectral invariants and a non-vanishing result of twisted Seiberg- Witten Floer homology. This is joint work with Dan Cristofaro-Gardiner and Rohil Prasad. |
|
Speaker: An Huang, Brandeis UniversityTitle: 12/16/2021 Interdisciplinary Science SeminarVenue: VirtualTitle: Quadratic reciprocity from a family of adelic conformal field theories Abstract: We consider a deformation of the 2d free scalar field action by raising the Laplacian to a positive real power. It turns out that the resulting non-local generalized free action is invariant under two commuting actions of the global conformal symmetry algebra, although it’s no longer invariant under the local conformal symmetry algebra. Furthermore, there is an adelic version of this family of global conformal field theories, parametrized by the choice of a number field, together with a Hecke character. Tate’s thesis plays an important role here in calculating Green’s functions of these theories, and in ensuring the adelic compatibility of these theories. In particular, the local L-factors… |
|
Speaker: Michael Douglas, Simons Center/CMSATitle: 12/9/21 Interdisciplinary Science SeminarVenue: VirtualTitle: Numerical Higher Dimensional Geometry Abstract: In 1977, Yau proved that a Kahler manifold with zero first Chern class admits a Ricci flat metric, which is uniquely determined by certain “moduli” data. These metrics have been very important in mathematics and in theoretical physics, but despite much subsequent work we have no analytical expressions for them. But significant progress has been made on computing numerical approximations. We give an introduction (not assuming knowledge of complex geometry) to these problems and describe these methods. |
|
Speaker: Ryosuke Takahashi, National Cheng Kung UniversityTitle: 12/2/2021 Interdisciplinary Science SeminarVenue: VirtualTitle: Polyhomogeneous expansions and Z/2-harmonic spinors branching along graphs Abstract: In this talk, we will first reformulate the linearization of the moduli space of Z/2-harmonic spinorsv branching along a knot. This formula tells us that the kernel and cokernel of the linearization are isomorphic to the kernel and cokernel of the Dirac equation with a polyhomogeneous boundary condition. In the second part of this talk, I will describe the polyhomogenous expansions for the Z/2-harmonic spinors branching along graphs and formulate the Dirac equation with a suitable boundary condition that can describe the perturbation of graphs with some restrictions. This is joint work with Andriy Haydys and Rafe Mazzeo. |
|
Speaker: Matteo Parisi, CMSA/IASTitle: 11/18/2021 Interdisciplinary Science SeminarVenue: VirtualTitle: Amplituhedra, Scattering Amplitudes and Triangulations Abstract: In this talk I will discuss about Amplituhedra – generalizations of polytopes inside the Grassmannian – recently introduced by physicists as new geometric constructions encoding interactions of elementary particles in certain Quantum Field Theories. In particular, I will explain how the problem of finding triangulations of Amplituhedra is connected to computing scattering amplitudes of N=4 super Yang-Mills theory. Triangulations of polygons are encoded in the associahedron studied by Stasheff in the sixties; in the case of polytopes, triangulations are captured by secondary polytopes constructed by Gelfand et al. in the nineties. Whereas a “secondary” geometry describing triangulations of Amplituhedra is still not known, and we pave the way for such studies. We will… |
|
Speaker: Lvzhou Chen, University of Texas at AustinTitle: 11/11/21 Interdisciplinary Science SeminarVenue: VirtualTitle: The Kervaire conjecture and the minimal complexity of surfaces Abstract: We use topological methods to solve special cases of a fundamental problem in group theory, the Kervaire conjecture. The conjecture asserts that, for any nontrivial group G and any element w in the free product G*Z, the quotient (G*Z)/<<w>> is still nontrivial. We interpret this as a problem of estimating the minimal complexity (in terms of Euler characteristic) of surfaces in HNN extensions. This gives a conceptually simple proof of Klyachko’s theorem that confirms the Kervaire conjecture for any G torsion-free. I will also explain new results obtained using this approach. |
|
Speaker: Qifeng Chen, The Hong Kong University of Science and TechnologyTitle: 11/4/21 CMSA Interdisciplinary Science SeminarVenue: VirtualTitle: Exploring Invertibility in Image Processing and Restoration Abstract: Today’s smartphones have enabled numerous stunning visual effects from denoising to beautification, and we can share high-quality JPEG images easily on the internet, but it is still valuable for photographers and researchers to keep the original raw camera data for further post-processing (e.g., retouching) and analysis. However, the huge size of raw data hinders its popularity in practice, so can we almost perfectly restore the raw data from a compressed RGB image and thus avoid storing any raw data? This question leads us to design an invertible image signal processing pipeline. Then we further explore invertibility in other image processing and restoration tasks, including image compression, reversible image conversion (e.g., image-to-video… |
|
Speaker: Jie Yang, Delft University of TechnologyTitle: ARCH: Know What Your Machine Doesn’t KnowVenue: VirtualSpeaker: Jie Yang, Delft University of Technology Title: ARCH: Know What Your Machine Doesn’t Know Abstract: Despite their impressive performance, machine learning systems remain prohibitively unreliable in safety-, trust-, and ethically sensitive domains. Recent discussions in different sub-fields of AI have reached the consensus of knowledge need in machine learning; few discussions have touched upon the diagnosis of what knowledge is needed. In this talk, I will present our ongoing work on ARCH, a knowledge-driven, human-centered, and reasoning-based tool, for diagnosing the unknowns of a machine learning system. ARCH leverages human intelligence to create domain knowledge required for a given task and to describe the internal behavior of a machine learning system; it infers the missing or incorrect knowledge of… |
|
Speaker: Colin Guillarmou, CNRS/Univ. Paris SaclayTitle: 10/21/2021 Interdisciplinary Science SeminarVenue: VirtualTitle: Mathematical resolution of the Liouville conformal field theory. Abstract: The Liouville conformal field theory is a well-known beautiful quantum field theory in physics describing random surfaces. Only recently a mathematical approach based on a well-defined path integral to this theory has been proposed using probability by David, Kupiainen, Rhodes, Vargas. Many works since the ’80s in theoretical physics (starting with Belavin-Polyakov-Zamolodchikov) tell us that conformal field theories in dimension 2 are in general « Integrable », the correlations functions are solutions of PDEs and can in principle be computed explicitely by using algebraic tools (vertex operator algebras, representations of Virasoro algebras, the theory of conformal blocks). However, for Liouville Theory this was not done at the mathematical… |
|
Speaker: Ian Gemp, DeepMindTitle: 10/14/2021 Interdisciplinary Science SeminarVenue: VirtualTitle: D3C: Reducing the Price of Anarchy in Multi-Agent Learning Abstract: In multi-agent systems the complex interaction of fixed incentives can lead agents to outcomes that are poor (inefficient) not only for the group but also for each individual agent. Price of anarchy is a technical game theoretic definition introduced to quantify the inefficiency arising in these scenarios– it compares the welfare that can be achieved through perfect coordination against that achieved by self-interested agents at a Nash equilibrium. We derive a differentiable upper bound on a price of anarchy that agents can cheaply estimate during learning. Equipped with this estimator agents can adjust their incentives in a way that improves the efficiency incurred at a Nash equilibrium. Agents adjust… |