Analyzing the Large System Limit of Stochastic Proximal Gradient Methods for Online Estimation

Abstract

The objective of this project is to analyze the exact dynamics of online stochastic proximal gradient methods in the large-system limit. The theoretical analysis will lead to new insights and more principled methods to select the various parameters of the algorithms to achieve optimal trade-offs between estimation accuracy and computational complexity. This research will focus on (1) characterizing the asymptotic dynamics of online stochastic proximal gradient methods via deterministic systems of ODEs; (2) performing a more refined analysis in terms of the density evolutions of the error vectors; and (3) creating principled methods to optimize the various parameters of the algorithms to achieve optimal trade-offs between estimation accuracy and computational complexity.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1610265

Entities

People

  • Yue M. Lu

Organizations

  • Army Contracting Command
  • Harvard University
  • United States Army

Tags

Readers

  • Calculus or Mathematical Analysis
  • Distributed Systems and Data Platform Development
  • Operations Research