Sample-Efficient Nonconvex Optimization Algorithms for Machine Learning

Zoom link

Passcode:  292816

Abstract:  Nonconvex optimization plays a central role in modern machine learning. How to design data-efficient optimization algorithms that have a low sample complexity while enjoying a fast convergence at the same time remains a pressing and challenging research question in machine learning. My research aims to answer this question from two facets: providing the theoretical understanding of optimization algorithms; and developing new algorithms with strong empirical performance in a principled way.

In this talk, I will discuss the sample efficiency of stochastic gradient-based algorithms for solving nonconvex optimization problems. I will first introduce algorithms based on new variance reduction techniques that achieve improved sample efficiency over existing first-order optimization methods. Then I will show that these variance reduction techniques can be used to develop sample-efficient algorithms for policy optimization problems in reinforcement learning. Lastly, I will also describe multiple applications of efficient machine learning algorithms to a diverse set of interdisciplinary areas and some ongoing and future directions.

Bio:  Pan Xu is a Ph.D. candidate in the Department of Computer Science at the University of California, Los Angeles. His research spans the areas of machine learning, optimization, reinforcement learning, and high-dimensional statistics, with a focus on the development and improvement of large-scale nonconvex optimization algorithms for machine learning applications. Pan’s work is published in top machine learning conferences and journals such as ICML, NeurIPS, ICLR, AISTATS, and JMLR. He has received awards including Presidential Fellowship in Data Science (2016) from the University of Virginia, and Rising Stars in Data Science (2021) from the University of Chicago.