Most non-equilibrium dynamics are not driven by a gradient of potential, and thus properties of steady states can not be analytically packed into an explicit function like a Boltzmann distribution. Here, we take a new angle that only slow points in phase space are considered, and an optimization-based equilibrium measure can thus be constructed. Our theory reveals the continuous nature of chaos transition in a neural network model, and further identifies a response order parameter peaked at the transition, surprisingly consistent with cortical observation of conscious brain states. Our framework thus opens a new route to analyze non-equilibrium steady states.

报告人简介

黄海平，中山大学教授，博士生导师，2011年获得中科院理论物理所博士学位，随后在香港科技大学物理系、东京工业大学计算智能系、日本理化学研究所(RIKEN) 脑科学中心从事统计物理与机器学习、神经计算交叉的基础理论研究，曾获得JSPS博士后资助、RIKEN 杰出研究奖，国家自然科学优秀青年基金等，著有英文专著《神经网络的统计力学》（德国施普林格出版社和高等教育出版社2022年联合出版）

邀请人

周海军 研究员