A Mean-field Optimal Control Formulation of Deep Learning

Colloquia & Seminars

A Mean-field Optimal Control Formulation of Deep Learning
Date/Time:14 Mar 2019 16:00 Venue: S17 #05-12 SR4 Speaker: Li Qianxiao, National University of Singapore A Mean-field Optimal Control Formulation of Deep Learning In this talk, we discuss formulating, through a continuous-time approximation, deep supervised learning as a mean-field optimal control problem. We prove necessary conditions for optimality in the form of a mean-field Pontryagin’s maximum principle, as well as global characterizations of optimality using Hamilton-Jacobi-Bellman equations. Interesting learning algorithms derived from this approach will also be discussed. Add to calendar: