基于属性引导的大型语言模型持续学习
Attribution-Guided Continual Learning for Large Language Models
AI总结：
本文提出基于属性引导的持续学习框架，通过估计每个Transformer层中参数的重要性，调节梯度以保留旧任务知识，实验表明在持续学习基准上优于基线方法。
摘要：
大型语言模型（LLMs）在持续学习中常出现灾难性遗忘：在依次学习新任务后，其在早期任务上的表现变差。现有方法通过数据回放、参数冻结或正则化来缓解灾难性遗忘，但这些方法缺乏对LLM内部知识分布语义意识。为此，我们提出一种属性引导的持续微调框架。我们的方法估计每个Transformer层中任务特定的参数重要性，并利用这些分数调节梯度。对先前任务重要的参数接收较小的更新，而相关性较低的参数保持可塑性以学习新任务。在持续学习基准上的实验表明，我们的方法在保持旧任务性能的同时，能够维持在新任务上的竞争力。
英文摘要：
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.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
