AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection

TL;DR

AnaCP introduces a novel method for class-incremental learning that preserves feature adaptation without gradient updates, significantly reducing catastrophic forgetting and matching joint training performance.

Contribution

AnaCP is the first approach to enable incremental feature adaptation in PTM-based CIL without gradient updates, improving performance and reducing forgetting.

Findings

AnaCP outperforms existing baselines in CIL tasks.

AnaCP achieves accuracy comparable to joint training.

AnaCP effectively eliminates catastrophic forgetting.

Abstract

This paper studies the problem of class-incremental learning (CIL), a core setting within continual learning where a model learns a sequence of tasks, each containing a distinct set of classes. Traditional CIL methods, which do not leverage pre-trained models (PTMs), suffer from catastrophic forgetting (CF) due to the need to incrementally learn both feature representations and the classifier. The integration of PTMs into CIL has recently led to efficient approaches that treat the PTM as a fixed feature extractor combined with analytic classifiers, achieving state-of-the-art performance. However, they still face a major limitation: the inability to continually adapt feature representations to best suit the CIL tasks, leading to suboptimal performance. To address this, we propose AnaCP (Analytic Contrastive Projection), a novel method that preserves the efficiency of analytic classifiers while enabling incremental feature adaptation without gradient-based training, thereby eliminating the CF caused by gradient updates. Our experiments show that AnaCP not only outperforms existing baselines but also achieves the accuracy level of joint training, which is regarded as the upper bound of CIL.