Rethinking the theoretical foundation of reinforcement learning

Abstract: Given two candidate functions, can we identify which one is the true value function of a large Markov decision process (MDP), given a "benign" dataset? Trivial as it might seem, a version of the question was open for 20+ years in reinforcement learning (RL), and the core difficulties are intimately related to the training instability of modern deep RL. In this talk, I will argue that ...

Speaker：Nan Jiang (University of Illinois at Urbana-Champaign)

Quantification of ergodicity for Hamilton–Jacobi equations in a dynamic random environment

Abstract: Motivated by the tensionless KPZ equation (a model of stochastic growth), we consider quantitative large-time averages for Hamilton–Jacobi equations in a dynamic random environment that is stationary ergodic and has finite-range dependence in time. In this talk, we establish, up to slowly varying factors, convergence rates with exponent 1/2 for the large-time averages of both the sol...

Speaker：Prof. GUO Xiaoqin (郭小钦), University of Cincinnati

Toward Convergent Adam-Type Methods

Abstract:Adam is one of the most widely used optimizers in modern machine learning, but its convergence theory remains incomplete, even in relatively simple convex settings. A central difficulty is that Adam tightly couples momentum and adaptive preconditioning, which hides the dissipative structure needed for a Lyapunov analysis.In this talk, I will describe a Lyapunov-guided approach to Adam-...

Speaker：Long Chen (University of California, Irvine)

A deep learning method for computing the spectrum of the Schrödinger operator: numerics and analysis

Abstract: We present a novel deep learning method for computing eigenstates of the Schrödinger operator. Our approach combines a novel loss function with an innovative neural network architecture that incorporates a-prior knowledge of the problem. These improvements enable the method to handle both high-dimensional problems and problems posed on irregular bounded domains. We successfully compu...

Speaker：Ming Pingbing (CAS)

Critical interfaces and SLE curves: power law rate of convergence

Abstract:Many two‑dimensional lattice models from statistical physics are believed to exhibit universal scaling limits at criticality, described by Schramm–Loewner Evolution (SLE). While convergence to SLE has been established in several celebrated cases, much less is known about how fast this convergence occurs. In this talk, I will discuss recent progress on quantitative convergence results...

Speaker：Ilia Binder (University of Toronto)