arXiv cs.AI· Md Anwar Hossen, Fatema Siddika, Juan Pablo Munoz, Tanya Roosta, Ali Jannesari·· 2 天前AI 评分22
CRAFT：面向持续学习的遗忘感知干预适配框架
CRAFT: Forgetting-Aware Intervention-Based Adaptation for Continual Learning
AI 导读
研究者提出持续学习框架 CRAFT，冻结 LLM 权重，改为在隐藏表示上学习低秩干预。相似任务共享同一干预以促进知识迁移，任务相似度由各任务对模型输出分布的改变程度判定；共享干预内用 KL 惩罚约束新任务贴近既有行为以限制遗忘，最深层的小型任务私有干预则保留后续任务无法覆盖的专属容量。在多个基准和模型规模上，CRAFT 整体表现优于强 LoRA 方法并减少遗忘，且对任务顺序保持稳健。
正文 · 原文
Abstract:Large language models (LLMs) can learn new tasks through fine-tuning, but learning tasks sequentially often leads to catastrophic forgetting of earlier ones. This paper proposes CRAFT, a continual learning framework that keeps model weights frozen and instead learns low-rank interventions on hidden representations. Similar tasks share an intervention to help knowledge transfer, with similarity judged by how each task alters the model's output distribution rather than by a learned gating network. Within a shared intervention, a Kullback-Leibler (KL) penalty keeps each new task close to the intervention's existing behavior, which limits forgetting of earlier tasks, while a small, task-private intervention in the deepest layers preserves dedicated capacity that later tasks cannot overwrite. The learned interventions also make it possible to identify the task at test time, so task identity labels are not needed. Across multiple benchmarks and model scales, CRAFT improves overall performance and reduces forgetting compared with strong LoRA-based methods, while remaining robust to task ordering.
| Comments: | 21 pages |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.05732 [cs.LG] |
| (or arXiv:2605.05732v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.05732 arXiv-issued DOI via DataCite |
Submission history
From: Md Anwar Hossen [view email]
[v1]
Thu, 7 May 2026 06:24:13 UTC (100 KB)
[v2]
Fri, 8 May 2026 04:10:34 UTC (98 KB)
[v3]
Wed, 14 Oct 2026 08:47:15 UTC (177 KB)
来源：arXiv cs.AI · arxiv.org