用机制分析指导大模型持续学习，减少遗忘同时保持新任务适应力。
Attribution-Guided Continual Learning for Large Language Models
- 基于层间重要性传播，识别关键参数与可更新参数
- 在持续学习中约束重要参数更新，减少灾难性遗忘
- 适合需要长期增量学习的大模型应用
大语言模型在持续学习中常出现灾难性遗忘：顺序学习新任务后，旧任务性能下降。现有方法通过数据重放、参数冻结或正则化缓解遗忘，但缺乏对模型内部机制的理解，无法区分哪些参数存储了旧知识，哪些可安全更新。为此，我们提出基于归因引导的持续微调框架，利用层间重要性传播（LRP）评估参数在模型内部计算过程中的重要性。在持续学习过程中，关键参数受控更新幅度较小，而无关参数仍可自由调整以学习新任务。大量实验表明，相比基线方法，本方案有效减少遗忘并保持对新任务的适应能力，凸显机制归因在大模型持续微调中的价值。
原文摘要 · Abstract (English)
Large language models (LLMs) often suffer from catastrophic forgetting in continual learning: after learning new tasks sequentially, they perform worse on earlier tasks. Existing methods mitigate catastrophic forgetting by data replay, parameter freezing, or regularization. However, these methods lack understanding of LLM mechanisms and cannot distinguish which parameters store important knowledge from previous tasks and which parameters can be updated for new tasks. To address this, we propose the attribution-guided continual fine-tuning framework that leverages Layer-wise Relevance Propagation (LRP) to estimate parameter importance based on the internal computational process of LLMs. During continual learning, parameters critical to previous tasks are constrained to receive smaller updates, while less relevant parameters remain available for learning new tasks. Extensive experiments show that, compared with baseline methods, our approach reduces catastrophic forgetting while preserving adaptability to new tasks, highlighting the value of mechanistic attribution for continual fine-tuning of LLMs.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。