Regularization Shortcomings for Continual Learning

TL;DR

This paper critically examines regularization-based continual learning methods, revealing their inability to distinguish classes across tasks in class-incremental scenarios, with implications for reinforcement learning and pre-trained models.

Contribution

It provides a theoretical analysis demonstrating the limitations of regularization approaches in continual learning, supported by experiments and practical implications.

Findings

Regularization methods fail to discriminate classes across tasks in class-incremental learning.

This shortcoming affects continual reinforcement learning and pre-trained models.

Understanding these limitations can guide more effective continual learning strategies.

Abstract

In most machine learning algorithms, training data is assumed to be independent and identically distributed (iid). When it is not the case, the algorithm's performances are challenged, leading to the famous phenomenon of catastrophic forgetting. Algorithms dealing with it are gathered in the Continual Learning research field. In this paper, we study the regularization based approaches to continual learning and show that those approaches can not learn to discriminate classes from different tasks in an elemental continual benchmark: the class-incremental scenario. We make theoretical reasoning to prove this shortcoming and illustrate it with examples and experiments. Moreover, we show that it can have some important consequences on continual multi-tasks reinforcement learning or in pre-trained models used for continual learning. We believe that highlighting and understanding the shortcomings of regularization strategies will help us to use them more efficiently.