CBA: Improving Online Continual Learning via Continual Bias Adaptor

TL;DR

This paper introduces the Continual Bias Adaptor (CBA), a module that enhances online continual learning by adapting to distribution shifts, improving knowledge retention without extra testing costs.

Contribution

The paper proposes a novel CBA module that mitigates catastrophic distribution shifts in online continual learning, with theoretical analysis and extensive empirical validation.

Findings

CBA effectively reduces forgetting in continual learning.

CBA achieves superior performance on multiple benchmarks.

The method introduces no additional cost during testing.

Abstract

Online continual learning (CL) aims to learn new knowledge and consolidate previously learned knowledge from non-stationary data streams. Due to the time-varying training setting, the model learned from a changing distribution easily forgets the previously learned knowledge and biases toward the newly received task. To address this problem, we propose a Continual Bias Adaptor (CBA) module to augment the classifier network to adapt to catastrophic distribution change during training, such that the classifier network is able to learn a stable consolidation of previously learned tasks. In the testing stage, CBA can be removed which introduces no additional computation cost and memory overhead. We theoretically reveal the reason why the proposed method can effectively alleviate catastrophic distribution shifts, and empirically demonstrate its effectiveness through extensive experiments based on four rehearsal-based baselines and three public continual learning benchmarks.