Mind the Interference: Retaining Pre-trained Knowledge in Parameter Efficient Continual Learning of Vision-Language Models

TL;DR

This paper introduces DIKI, a novel framework for continual learning in vision-language models that preserves pre-trained knowledge efficiently by avoiding interference, enabling better adaptation to diverse tasks with minimal parameter updates.

Contribution

The study proposes a distribution-aware, interference-free knowledge integration method that retains pre-trained VLM knowledge during continual learning, reducing computational costs and parameter updates.

Findings

Outperforms state-of-the-art with only 0.86% of parameters trained

Requires significantly less training time than existing methods

Effectively preserves zero-shot capabilities across tasks

Abstract

This study addresses the Domain-Class Incremental Learning problem, a realistic but challenging continual learning scenario where both the domain distribution and target classes vary across tasks. To handle these diverse tasks, pre-trained Vision-Language Models (VLMs) are introduced for their strong generalizability. However, this incurs a new problem: the knowledge encoded in the pre-trained VLMs may be disturbed when adapting to new tasks, compromising their inherent zero-shot ability. Existing methods tackle it by tuning VLMs with knowledge distillation on extra datasets, which demands heavy computation overhead. To address this problem efficiently, we propose the Distribution-aware Interference-free Knowledge Integration (DIKI) framework, retaining pre-trained knowledge of VLMs from a perspective of avoiding information interference. Specifically, we design a fully residual mechanism to infuse newly learned knowledge into a frozen backbone, while introducing minimal adverse impacts on pre-trained knowledge. Besides, this residual property enables our distribution-aware integration calibration scheme, explicitly controlling the information implantation process for test data from unseen distributions. Experiments demonstrate that our DIKI surpasses the current state-of-the-art approach using only 0.86% of the trained parameters and requiring substantially less training time. Code is available at: https://github.com/lloongx/DIKI .