Efficient Online Learning with Memory via Frank-Wolfe Optimization: Algorithms with Bounded Dynamic Regret and Applications to Control

By Hongyu Zhou et al

Published on March 31, 2023

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. Problem Formulation
4. Meta-OFW Algorithm for OCO-M

Summary

This paper introduces an algorithm for Online Convex Optimization with Memory (OCO-M) that minimizes dynamic regret and enables projection-free online learning. The algorithm is applied to online control of linear time-varying systems. It addresses challenges such as adapting to dynamic environments, capturing the effect of past decisions, and fast decision-making. The algorithm builds on the Online Frank-Wolfe (OFW) and Hedge algorithms. The paper also presents technical contributions and numerical evaluations demonstrating the algorithm's efficiency and performance.