Roles of Chaos and Noise in Learning Algorithms Across Physical Systems

Noise and chaos are ubiquitous in nature and are usually thought of as detrimental to information processing. In this talk, we turn to physical systems that can coexist with, and sometimes harness, noise and chaos for information processing. First, I will present our recent computational neuroscience findings showing how cortical networks can self-organize to operate, without parameter fine-tuning, near the “edge of chaos” - a desirable dynamical regime robust to noise yet flexible enough for temporal information processing. Using new analytical tools from sparse non-Hermitian random matrix theory, we show that this robust criticality emerges naturally from an inhibition-dominated network with heterogeneous synaptic timescales, which help maintain reliable temporal memory in noisy conditions.

 Secondly, we turn to population learning, where self-interested agents optimize their own rewards (reinforcement learning) in a game-theoretic setting. We show how self-interested individuals’ decisions can drive population dynamics in which chaos and unpredictability provably emerge, yet time-average macroscopic order residing at social optima (Nash equilibria) persists. This emergence of macroscopic order mirrors thermodynamics, where molecular chaos averages out to steady pressure, except now the passive particles are active decision-making programs chasing their own goals in their own ways. Together, these results outline design principles for learning algorithms that coexist with, and sometimes harness, unpredictability as a computational resource, which could perhaps inspire noise-resilient quantum learning algorithms.

Thiparat received his Ph.D. in theoretical physics under the supervision of David R. Nelson from Harvard University, focusing on non-equilibrium statistical mechanics of evolutionary processes and biological pattern formation in disordered media. Then, he explored the intersection of statistical mechanics, machine learning, and neuroscience as a Postdoctoral Scholar at Singapore University of Technology and Design (SUTD). He is currently an Assistant Professor of Physics at Chulalongkorn University, where he leads the Chula Intelligent and Complex Systems Center of Excellence. His research interests include quantum information science, statistical mechanics, and artificial intelligence from theoretical and computational perspectives. He received Chulalongkorn University’s 2025 Outstanding Researcher Award and 2024 Outstanding Junior Researcher Award in Physical Sciences and Mathematics, and led quantum simulations and algorithms project under Thailand Quantum Flagship Research Program, from 2020 to present. He also holds visiting scholar appointments at institutions including EPFL (Switzerland), and KITP (Santa Barbara).