Semantic-Guided Dynamic Sparsification for Pre-Trained Model-based Class-Incremental Learning

TL;DR

This paper introduces SGDS, a novel method for class-incremental learning that dynamically guides activation space to improve knowledge transfer and reduce interference, achieving state-of-the-art results.

Contribution

SGDS is a new approach that proactively shapes activation subspaces via targeted sparsification, enhancing plasticity and knowledge sharing in pre-trained models for CIL.

Findings

SGDS outperforms existing methods on multiple benchmarks.

It effectively balances knowledge transfer and interference prevention.

Demonstrates significant improvements in class-incremental learning tasks.

Abstract

Class-Incremental Learning (CIL) requires a model to continually learn new classes without forgetting old ones. A common and efficient solution freezes a pre-trained model and employs lightweight adapters, whose parameters are often forced to be orthogonal to prevent inter-task interference. However, we argue that this parameter-constraining method is detrimental to plasticity. To this end, we propose Semantic-Guided Dynamic Sparsification (SGDS), a novel method that proactively guides the activation space by governing the orientation and rank of its subspaces through targeted sparsification. Specifically, SGDS promotes knowledge transfer by encouraging similar classes to share a compact activation subspace, while simultaneously preventing interference by assigning non-overlapping activation subspaces to dissimilar classes. By sculpting class-specific sparse subspaces in the activation space, SGDS effectively mitigates interference without imposing rigid constraints on the parameter space. Extensive experiments on various benchmark datasets demonstrate the state-of-the-art performance of SGDS.