Adapt before Continual Learning

TL;DR

This paper introduces ACL, a pre-training adaptation framework that refines models before continual learning, improving the stability-plasticity balance and enhancing performance across benchmarks.

Contribution

It proposes a novel plug-and-play adaptation phase for PTMs before CL, addressing stability-plasticity trade-off more effectively than existing methods.

Findings

ACL improves CL performance across multiple benchmarks.

Refinement aligns embeddings with class prototypes.

Significant gains over baseline methods.

Abstract

Continual Learning (CL) seeks to enable neural networks to incrementally acquire new knowledge (plasticity) while retaining existing knowledge (stability). Although pre-trained models (PTMs) have provided a strong foundation for CL, existing approaches face a fundamental challenge in balancing these two competing objectives. Current methods typically address stability by freezing the PTM backbone, which severely limits the model's plasticity, particularly when incoming data distribution diverges largely from the pre-training data. Alternatively, sequentially fine-tuning the entire PTM can adapt to new knowledge but often leads to catastrophic forgetting, highlighting the critical stability-plasticity trade-off in PTM-based CL. To address this limitation, we propose Adapting PTMs before the core CL} process (ACL), a novel framework that introduces a plug-and-play adaptation phase prior to learning each new task. During this phase, ACL refines the PTM backbone by aligning embeddings with their original class prototypes while distancing them from irrelevant classes. This mechanism theoretically and empirically demonstrates desirable balance between stability and plasticity, significantly improving CL performance across benchmarks and integrated methods. Code is available at https://github.com/byyx666/ACL_code.