An Investigation of Replay-based Approaches for Continual Learning

TL;DR

This paper empirically evaluates replay-based methods for continual learning, highlighting the importance of sample selection strategies and showing that simple rehearsal approaches can outperform more complex recent methods.

Contribution

It provides a comprehensive comparison of replay-based approaches, including novel naive rehearsal methods, and analyzes the impact of sample selection on performance.

Findings

Sample selection strategies significantly affect performance.

Naive rehearsal approaches can outperform state-of-the-art methods.

Performance varies greatly among different replay techniques.

Abstract

Continual learning (CL) is a major challenge of machine learning (ML) and describes the ability to learn several tasks sequentially without catastrophic forgetting (CF). Recent works indicate that CL is a complex topic, even more so when real-world scenarios with multiple constraints are involved. Several solution classes have been proposed, of which so-called replay-based approaches seem very promising due to their simplicity and robustness. Such approaches store a subset of past samples in a dedicated memory for later processing: while this does not solve all problems, good results have been obtained. In this article, we empirically investigate replay-based approaches of continual learning and assess their potential for applications. Selected recent approaches as well as own proposals are compared on a common set of benchmarks, with a particular focus on assessing the performance of different sample selection strategies. We find that the impact of sample selection increases when a smaller number of samples is stored. Nevertheless, performance varies strongly between different replay approaches. Surprisingly, we find that the most naive rehearsal-based approaches that we propose here can outperform recent state-of-the-art methods.