Adaptive Adapter Routing for Long-Tailed Class-Incremental Learning

TL;DR

This paper introduces APART, a novel exemplar-free method leveraging adaptive adapter routing in pre-trained models to address long-tailed class-incremental learning, effectively mitigating forgetting and class imbalance without retraining classifiers.

Contribution

The paper proposes a new adapter-based framework with adaptive routing and adapter pools for LTCIL, avoiding retraining classifiers and improving handling of minority classes.

Findings

APART outperforms existing methods on benchmark datasets.

It effectively mitigates catastrophic forgetting.

The approach enhances minority class recognition.

Abstract

In our ever-evolving world, new data exhibits a long-tailed distribution, such as e-commerce platform reviews. This necessitates continuous model learning imbalanced data without forgetting, addressing the challenge of long-tailed class-incremental learning (LTCIL). Existing methods often rely on retraining linear classifiers with former data, which is impractical in real-world settings. In this paper, we harness the potent representation capabilities of pre-trained models and introduce AdaPtive Adapter RouTing (APART) as an exemplar-free solution for LTCIL. To counteract forgetting, we train inserted adapters with frozen pre-trained weights for deeper adaptation and maintain a pool of adapters for selection during sequential model updates. Additionally, we present an auxiliary adapter pool designed for effective generalization, especially on minority classes. Adaptive instance routing across these pools captures crucial correlations, facilitating a comprehensive representation of all classes. Consequently, APART tackles the imbalance problem as well as catastrophic forgetting in a unified framework. Extensive benchmark experiments validate the effectiveness of APART. Code is available at: https://github.com/vita-qzh/APART