DATA: Decomposed Attention-based Task Adaptation for Rehearsal-Free Continual Learning

2025-02-17

AI Summary

To address catastrophic forgetting in continual learning of large language models (LLMs), this paper proposes a data-free, decomposition-based attention task-adaptation framework. The method innovatively decouples task-specific and shared knowledge by introducing a collaborative high-rank and low-rank LoRA adapter mechanism. It further designs an attention-driven learnable weight generation module and a task-relevance-aware dynamic allocation strategy, augmented with stochastic parameter restoration during training to jointly enhance knowledge retention and model plasticity. Evaluated on three mainstream continual learning benchmarks, the approach achieves state-of-the-art performance, significantly mitigating forgetting while improving both rapid adaptation to new tasks and long-term knowledge preservation.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Continual learning (CL) is essential for Large Language Models (LLMs) to adapt to evolving real-world demands, yet they are susceptible to catastrophic forgetting (CF). While traditional CF solutions rely on expensive data rehearsal, recent rehearsal-free methods employ model-based and regularization-based strategies to address this issue. However, these approaches often neglect the model's plasticity, which is crucial to achieving optimal performance on newly learned tasks. Consequently, a key challenge in CL is striking a balance between preserving plasticity and mitigating CF. To tackle this challenge, we propose the $ extbf{D}$ecomposed $ extbf{A}$ttention-based $ extbf{T}$ask $ extbf{A}$daptation (DATA), which explicitly decouples and learns both task-specific and task-shared knowledge using high-rank and low-rank task adapters (e.g., LoRAs). For new tasks, DATA dynamically adjusts the weights of adapters of different ranks based on their relevance and distinction from previous tasks, allowing the model to acquire new task-specific skills while effectively retaining previously learned knowledge. Specifically, we implement a decomposed component weighting strategy comprising learnable components that collectively generate attention-based weights, allowing the model to integrate and utilize diverse knowledge from each DATA. Extensive experiments on three widely used benchmarks demonstrate that our proposed method achieves state-of-the-art performance. Notably, our approach significantly enhances model plasticity and mitigates CF by extending learnable components and employing stochastic restoration during training iterations.

Problem

• Balancing plasticity and catastrophic forgetting
• Adapting large models without data rehearsal
• Enhancing task-specific and shared knowledge learning

Innovation

• Decomposed Attention-based Task Adaptation
• High-rank and low-rank task adapters
• Learnable components weighting strategy