Schedule-Robust Online Continual Learning

TL;DR

This paper introduces schedule-robustness in continual learning, proposing a new method that maintains performance across arbitrary data presentation schedules, and demonstrating significant improvements on image classification benchmarks.

Contribution

The paper defines schedule-robustness for continual learning and presents a novel approach that learns a schedule-robust predictor adaptable with replay data.

Findings

Outperforms existing methods on CL benchmarks

Demonstrates robustness against arbitrary data schedules

Achieves large margin improvements in image classification

Abstract

A continual learning (CL) algorithm learns from a non-stationary data stream. The non-stationarity is modeled by some schedule that determines how data is presented over time. Most current methods make strong assumptions on the schedule and have unpredictable performance when such requirements are not met. A key challenge in CL is thus to design methods robust against arbitrary schedules over the same underlying data, since in real-world scenarios schedules are often unknown and dynamic. In this work, we introduce the notion of schedule-robustness for CL and a novel approach satisfying this desirable property in the challenging online class-incremental setting. We also present a new perspective on CL, as the process of learning a schedule-robust predictor, followed by adapting the predictor using only replay data. Empirically, we demonstrate that our approach outperforms existing methods on CL benchmarks for image classification by a large margin.