We discuss an analogue of the Goldbach conjecture for polynomials with coefficients in semidomains (i.e., subsemirings of an integral domain).

Empirical evidence suggests that for a variety of overparameterized nonlinear models, most notably in neural network training, the growth of the loss around a minimizer strongly impacts its performance. Flat minima -- those around which the loss grows slowly -- appear to generalize well. This work takes a step towards understanding this phenomenon by focusing on the simplest class of overparameterized nonlinear models: those arising in low-rank matrix recovery. We analyze overparameterized matrix and bilinear sensing, robust PCA, covariance matrix estimation, and single hidden layer neural networks with quadratic activation functions. In all cases, we show that flat minima, measured by the trace of the Hessian, exactly recover the ground truth under standard statistical assumptions. For matrix completion, we establish weak recovery, although empirical evidence suggests exact recovery holds here as well.

The q-circular system is a tuple of non-commutative random variables (operators with some state) which interpolate independent standard complex Gaussian random variables (q=1) in classical probability and freely independent circular random variables (q=0) in free probability. One of the interesting results on q-deformed probability is that -1<q<1 case has similar properties to free case (q=0). Haagerup inequality is one of such properties, which was originally proved for generators of free groups with respect to the left regular representation.    

 In this talk, I will explain the strong version of Haagerup inequality for the q-circular system, which was originally proved by Kemp and Speicher for q=0. This talk is based on a joint project with T. Kemp.

This talk discusses the matrix Moment-SOS hierarchy for solving polynomial matrix optimization problems. We first establish the finite convergence of this hierarchy under the Archimedean property, provided the nondegeneracy condition, strict complementarity condition, and second-order sufficient condition hold at every minimizer. Furthermore, we also prove that every minimizer of the moment relaxation must exhibit a flat truncation when the relaxation order is sufficiently large.

Quantum interaction models discussed here are the (asymmetric) quantum Rabi model (QRM) and non-commutative harmonic oscillator (NCHO). The QRM is the most fundamental model describing the interaction between a photon and two-level atoms. The NCHO can be considered as a covering model of the QRM, and recently, the eigenvalue problems of NCHO and two-photon QRM (2pQRM) are shown to be equivalent. Spectral degeneracy can occur in models, but correspondingly there is a hidden symmetry relates geometrical nature described by hyperelliptic curves. In addition, the analytical formula for the heat kernel (propagator)/partition function of the QRM is described as a discrete path integral and gives the meromorphic continuation of its spectral zeta function (SZF). This discrete path integral can be interpreted to the irreducible decomposition of the infinite symmetric group $\mathfrak{S}_\infty$ naturally acting on $\mathbb{F}_2^\infty$, $\mathbb{F}_2$ being the binary field. Moreover, from the special values of the SZF of NCHO, an analogue of the Apéry numbers is naturally appearing, and their generating functions are, e.g., given by modular forms, Eichler integrals of a congruence subgroup. The talk overviews those above and present questions which are open.

[pre-talk at 3:00PM]

There is a special entropy quantity associated to the Gauss curvature flow which plays an important rule for the convergence of the flow. Similar entropy can also be defined for a class of generalized Gauss curvature flows, in particular for anisotropic flows. One crucial property is monotonicity of the associated entropy along the flow. Another is the fact that critical point of entropy associated to the anisotropic flow under volume constraint is a solution to the $L^p$-Minkowski problem. This provides a flow approach to the $L^p$-Minkowski problem. The main question is under what condition entropy can control the diameter, as to obtain non-collapsing estimate for the flow.  We will discuss the main steps of the approach, and open problems related to inhomogeneous type flows.

I will prove strict comparison of C* algebras associated to free groups and then use it to solve the C* version of Tarski's problem from 1945 in the negative. It is joint work with Amrutam, Gao and Patchell and another joint work with Schafhauser.

The key property of linear dispersive flows is that waves with different frequencies travel with different group velocities, which leads to the phenomena of dispersive decay. Nonlinear dispersive flows also allow for interactions of linear waves, and their long time behavior is determined by the balance of linear dispersion on one hand, and nonlinear effects on the other hand. 

The first goal of this talk will be to present a new set of conjectures which aim to describe the global well-posedness and the dispersive properties of solutions in the most difficult case when the nonlinear effects are dominant, assuming only small initial data. This covers many interesting physical models, yet, as recently as a few years ago, there was no clue even as to what one might reasonably expect. The second objective of the talk will be to describe some very recent results in this direction.  This is joint work with Mihaela Ifrim.

Deep learning has been wildly successful in practice and most state-of-the-art artificial intelligence systems are based on neural networks. Lacking, however, is a rigorous mathematical theory that adequately explains the amazing performance of deep neural networks. In this talk, I present a new mathematical framework that provides the beginning of a deeper understanding of deep learning. This framework precisely characterizes the functional properties of trained neural networks. The key mathematical tools which support this framework include transform-domain sparse regularization, the Radon transform of computed tomography, and approximation theory. This framework explains the effect of weight decay regularization in neural network training, the importance of skip connections and low-rank weight matrices in network architectures, the role of sparsity in neural networks, and explains why neural networks can perform well in high-dimensional problems. At the end of the talk we shall conclude with a number of open problems and interesting research directions.

This talk is based on work done in collaboration with Rob Nowak, Ron DeVore, Jonathan Siegel, Joe Shenouda, and Michael Unser.

Coq is a programming language and "proof assistant", where one can state and prove theorems which are checked for soundness by Coq itself. Looking at an example formalization of the hypernatural numbers, we'll explore what makes such a tool useful, interesting, and even fun! At the end of this talk attendees will hopefully have reasons to consider using Coq or similar tools themselves, and incidentally be able to construct a non-standard model of arithmetic (whenever the need arises).