Representation Finetuning for Continual Learning

2026-03-11

AI Summary

Existing parameter-efficient fine-tuning methods struggle to control representation drift in continual learning, rendering them susceptible to domain shift and catastrophic forgetting. To address this, this work proposes CoRe, a novel framework that shifts the fine-tuning paradigm from weight space to representation space. CoRe explicitly regulates representation updates by performing task-specific interventions within a low-rank linear subspace of hidden representations, thereby balancing stability for previously learned tasks with plasticity for new ones. Evaluated across multiple continual learning benchmarks, CoRe significantly outperforms current state-of-the-art approaches while maintaining high parameter efficiency and strong interpretability.

Technology Category

• Machine Learning: Representation Learning
• Computer Vision: Representation Learning for Vision
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

The world is inherently dynamic, and continual learning aims to enable models to adapt to ever-evolving data streams. While pre-trained models have shown powerful performance in continual learning, they still require finetuning to adapt effectively to downstream tasks. However, prevailing Parameter-Efficient Fine-Tuning (PEFT) methods operate through empirical, black-box optimization at the weight level. These approaches lack explicit control over representation drift, leading to sensitivity to domain shifts and catastrophic forgetting in continual learning scenarios. In this work, we introduce Continual Representation Learning (CoRe), a novel framework that for the first time shifts the finetuning paradigm from weight space to representation space. Unlike conventional methods, CoRe performs task-specific interventions within a low-rank linear subspace of hidden representations, adopting a learning process with explicit objectives, which ensures stability for past tasks while maintaining plasticity for new ones. By constraining updates to a low-rank subspace, CoRe achieves exceptional parameter efficiency. Extensive experiments across multiple continual learning benchmarks demonstrate that CoRe not only preserves parameter efficiency but also significantly outperforms existing state-of-the-art methods. Our work introduces representation finetuning as a new, more effective and interpretable paradigm for continual learning.

Problem

• continual learning
• representation drift
• catastrophic forgetting
• parameter-efficient fine-tuning
• domain shift

Innovation

• Continual Learning
• Representation Finetuning
• Low-Rank Subspace
• Parameter-Efficient Fine-Tuning
• Catastrophic Forgetting