演講題目：Distributed Incremental Methods for Finite-sum Optimization Problems

演   講   者：Prof. Hoi To Wai (Chinese University of Hong Kong)

時　　間：2019/1/28 (Mon.) 11:00~12:00

地　　點：清華大學台達館304室

ABSTRACT:

Finite sum optimization problems arise in many applications of machine learning and control systems. To handle them efficiently, one applies incremental method whose main idea is to divide the objective into batches and process the batches one-at-a-time. This talk presents our recent efforts on distributed optimization of using two variations of this popular technique.

The first part presents a curvature aided technique for incremental gradient methods, which uses second order information to provide improved tracking accuracy of the gradients. We develop a Stochastic Unbiased Curvature-Aided Gradient (SUCAG) method and analyze its convergence under a general asynchronous model of computation. We show that this model is compatible with a communication efficient implementation of distributed optimization based on random walk, and our result demonstrates that the convergence rate of SUCAG is robust to the graph topology.

The second part discusses the use of incremental methods in policy evaluation for multi-agent reinforcement learning. We demonstrate that the latter can be reformulated as a finite sum, primal dual optimization. Under some mild regularization conditions, we show that the latter can be solved efficiently using a distributed, incremental, and primal-dual optimization algorithm.

BIOGRAPHY  

Hoi-To Wai received his PhD degree from Arizona State University (ASU) in Electrical Engineering in Fall 2017, B. Eng. (with First Class Honor) and M. Phil. degrees in Electronic Engineering from The Chinese University of Hong Kong (CUHK) in 2010 and 2012, respectively. Since December 2018, he is an Assistant Professor in the Department of System Engineering & Engineering Management at his Alma Mater, CUHK. Previously, he has held research positions at ASU, UC Davis, Telecom ParisTech, Ecole Polytechnique, and MIT’s LIDS. Hoi-To's research interests are in the broad area of signal processing, machine learning and distributed optimization, with a focus of their applications to network science. His dissertation has received the 2017's Dean's Dissertation Award from the Ira A. Fulton Schools of Engineering of ASU and he is a recipient of a Best Student Paper Award at ICASSP 2018.