Week of November 6, 2022

Mon Nov 7, 2022
12:00pm - zoom - Probability and Analysis Webinar
Joseph Slote - (Caltech)
TBA

https://sites.google.com/view/paw-seminar/

4:00pm to 5:00pm - RH 306 - Applied and Computational Mathematics
Yuan Yao - (Hong Kong University of Science and Technology )
Robust Statistical Learning and Generative Adversarial Networks

Robust learning under Huber's contamination model has become an important topic in statistics and theoretical computer science. Statistically optimal procedures such as Tukey's median and other estimators based on depth functions are impractical because of their computational intractability. In this talk, we present an intriguing connection between f-GANs and various depth functions through the lens of f-Learning. Similar to the derivation of f-GANs, we show that these depth functions that lead to statistically optimal robust estimators can all be viewed as variational lower bounds of the total variation distance in the framework of f-Learning. This connection opens the door of computing robust estimators using tools developed for training GANs. In particular, we show in both theory and experiments that some appropriate structures of discriminator networks with hidden layers in GANs lead to statistically optimal robust location estimators for both Gaussian distribution and general elliptical distributions where first moment may not exist. Some applications are discussed on financial data analysis and robust denoising of Cryo-EM images.

Tue Nov 8, 2022
1:00pm to 2:00pm - RH 440R - Dynamical Systems
Alex Luna and Grigorii Monakov - (UC Irvine)
Dynamical systems via problem solving III: subshifts

This is a "problem solving session" in a series that we plan to continue this year, aimed at graduate students who would like to get familiar with some aspects of dynamical systems. This set of problems covers basic examples of symbolic dynamical systems, including subshifts of finite type etc.. No preliminary background is expected from the participants. Everybody is welcome!

1:00pm to 2:00pm - RH 306 - Mathematical Physics
Dan Borgnia - (UC Berkeley)
Projected Green’s Function Methods Applied to Quasi-Periodic Systems and the Dry Ten Martini Problem, 1

Projected Green’s Functions (pGf), Gxx(⍵), have long been used to describe the localization of quantum systems. More recently, pGf zeros have been used to determine physical observables of topological invariants in free-fermion systems, including topological obstructions to bulk localization and bulk-boundary correspondence. In this talk, I will discuss how these pGfs appear in transfer matrices and what their zeros can tell us about the solutions to transfer matrix equations – linking the localization and topological perspectives. Using these methods, we re-examine the almost-Matthieu operator and notice new guarantees on analytic regions of its resolvent.

Wed Nov 9, 2022
2:00pm to 3:00pm - 510R Rowland Hall - Combinatorics and Probability
Ludovic Stephan - (EPFL)
Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks

Despite the non-convex optimization landscape, over-parametrized shallow networks are able to achieve global convergence under gradient descent. The picture can be radically different for narrow networks, which tend to get stuck in badly-generalizing local minima. Here we investigate the cross-over between these two regimes in the high-dimensional setting, and in particular, investigate the connection between the so-called mean field/hydrodynamic regime and the seminal approach of Saad & Solla. Focusing on the case of Gaussian data, we study the interplay between the learning rate, the time scale, and the number of hidden units in the high-dimensional dynamics of stochastic gradient descent (SGD). Our work builds on a deterministic description of SGD in high-dimensions from statistical physics, which we extend and for which we provide rigorous convergence rates. https://arxiv.org/abs/2202.00293

Thu Nov 10, 2022
9:00am to 10:00am - Zoom - Inverse Problems
Arnold Kim - (UC Merced)
Tunably high-resolution synthetic aperture imaging

https://sites.uci.edu/inverse/

11:00am - RH 306 - Harmonic Analysis
Alexandros Eskenazis - (Sorbonne Université)
Dimension reduction for incompressible subsets of $\ell_p$

We present a Johnson-Lindenstrauss-type dimension reduction algorithm with additive error for incompressible subsets of $\ell_p$. The proof relies on a derandomized version of Maurey’s empirical method and a combinatorial idea of Ball.

1:00pm - RH 340N - Algebra
Daniel Rogalski - (UC San Diego)
Artin-Schelter regular algebras

What are the noncommutative rings that are most analogous to polynomial rings?  One class of such rings are the regular algebras first defined by Artin and Schelter in 1987.  Since then such algebras have been extensively studied.  I will give a survey of these interesting examples and some results about them.  I will also mention recent work of Manny Reyes and myself on the more general class of twisted Calabi-Yau algebras.

4:00pm - RH 192 - Differential Geometry
Lashi Bandara (note change in day and room) - (Brunel University London)
Boundary value problems for first-order elliptic operators with compact and noncompact boundary.

The index theorem for compact manifolds with boundary, established by Atiyah-Patodi-Singer in the mid-70s, is considered one of the most significant mathematical achievements of the 20th century. An important and curious fact is that local boundary conditions are topologically obstructed for index formulae and non-local  boundary conditions lie at the heart of this theorem. Consequently, this has inspired the study of boundary value problems for first-order elliptic differential operators by many different schools, with a class of induced  operators adapted to the boundary taking centre stage in formulating and understanding non-local boundary conditions.

That being said, much of this analysis has been confined to the situation when adapted boundary operators can be chosen self-adjoint. Dirac-type operators are the quintessential example. Nevertheless, natural geometric operators such as the Rarita-Schwinger operator on 3/2-spiniors, arising from physics in the study of the so-called Delta baryon, falls outside of this class. Analytically, this requires analysis beyond self-adjoint operators. In recent work with Bär,  the compact boundary case is handled for general first-order elliptic operators, using spectral theory to choose adapted boundary operators to be invertible bi-sectorial. The Fourier circle methods present in the self-adjoint analysis are replaced by the  bounded holomorphic functional calculus, coupled with pseudo-differential operator theory and semi-group techniques. This allows for a full understanding of the maximal domain of the interior operator as a bounded surjection to a space on the boundary of mixed Sobolev regularity, constructed from spectral projectors associated to the adapted boundary operator. Regularity and Fredholm extensions are also studied.

For the noncompact case, a preliminary trace theorem as well as regularity theory are handed by resorting to the case with compact boundary. This necessitates  deforming the coefficients of the interior operator in a compact neighbourhood. Therefore, even for Dirac-type operators, allowing for fully general symbols in the compact boundary case is paramount. Under slightly stronger geometric assumptions near the noncompact boundary (automatic  for the compact case) and  when the interior operator admits a self-adjoint adapted boundary operator, an upgraded trace theorem mirroring the compact setting is obtained. Importantly, there is no spectral assumptions other than self-adjointness on the adapted boundary operator. This, in particular, means that the spectrum of this operator can be the entire real line. Again, the primarily tool that is used in the analysis is the bounded holomorphic functional calculus.

4:00pm to 5:00pm - RH 306 - Mathematical Physics
Dan Borgnia - (UC Berkeley)
Projected Green’s Function Methods Applied to Quasi-Periodic Systems and the Dry Ten Martini Problem

Projected Green’s Functions (pGf), Gxx(⍵), have long been used to describe the localization of quantum systems. More recently, pGf zeros have been used to determine physical observables of topological invariants in free-fermion systems, including topological obstructions to bulk localization and bulk-boundary correspondence. In this talk, I will discuss how these pGfs appear in transfer matrices and what their zeros can tell us about the solutions to transfer matrix equations – linking the localization and topological perspectives. Using these methods, we re-examine the almost-Matthieu operator and notice new guarantees on analytic regions of its resolvent.