Studying Generalization on Memory-Based Methods in Continual Learning

TL;DR

This paper investigates how memory-based continual learning methods, while effective for in-distribution tasks, tend to overfit replay memory and perform poorly on out-of-distribution data, especially in linear classifiers.

Contribution

It reveals the limitations of memory-based methods in out-of-distribution generalization and highlights the tendency to learn spurious features, using a controlled benchmark environment.

Findings

Memory-based methods impair out-of-distribution generalization

Overfitting to replay memory occurs in these methods

Linear classifiers are most affected by this issue

Abstract

One of the objectives of Continual Learning is to learn new concepts continually over a stream of experiences and at the same time avoid catastrophic forgetting. To mitigate complete knowledge overwriting, memory-based methods store a percentage of previous data distributions to be used during training. Although these methods produce good results, few studies have tested their out-of-distribution generalization properties, as well as whether these methods overfit the replay memory. In this work, we show that although these methods can help in traditional in-distribution generalization, they can strongly impair out-of-distribution generalization by learning spurious features and correlations. Using a controlled environment, the Synbol benchmark generator (Lacoste et al., 2020), we demonstrate that this lack of out-of-distribution generalization mainly occurs in the linear classifier.