arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.36734cs.CL
收藏

CAMR:面向大语言模型持续学习的兼容性感知记忆重放

CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models

Beijing Zhongguancun Academy(北京中关村学院);Northeastern University(东北大学);Beihang University(北京航空航天大学);Renmin University of China(中国人民大学)

Jingpu Yang, Tianhao Li, Fengxian Ji, Yufeng Wang, Shuxin Zheng, Xiaoqing Zhang

AI总结:

提出CAMR自适应框架,结合兼容性检查与记忆门指导在线策略,在重放预算约束下优化大语言模型持续学习,在WikiBigEdit及四项TRACE任务中表现优于基线方法。

摘要:

持续学习要求大语言模型在获取新知识与能力的同时保留已学内容。微调与重放可支持知识获取与保留,但判断输入样本是否需要监督、应重放哪些历史样本仍是挑战。我们提出兼容性感知记忆重放(CAMR),这是一种自适应框架,可联合控制当前样本的监督与历史重放。CAMR将兼容性检查与学习到的记忆门结合,指导在线策略执行三类操作:结合通用重放学习输入样本、联合强化兼容相关样本、跳过对已验证掌握的重复样本的直接监督。该策略在每轮输入的重放预算约束下,从知识获取、保留及训练成本反馈中学习。我们在持续事实更新、跨任务与跨领域学习场景中,使用Qwen3-8B-Base评估CAMR。在WikiBigEdit数据集上,CAMR在3K至100K事实范围内的原始匹配度均优于所有评估方法,仅在3K事实时就达到93.66%,并在100K事实时超越全历史重放(LoRA)。在选定的四项TRACE任务中,CAMR均取得最高最终得分,且在涵盖金融问答、数学推理、代码生成与摘要的序列任务中,保持或提升了早期任务的性能。

英文摘要:

Continual learning requires large language models to acquire new knowledge and capabilities while retaining what they have learned. Fine-tuning and replay support acquisition and retention, but deciding whether incoming examples need supervision and which historical examples to revisit remains a challenge. We propose Compatibility-Aware Memory Replay (CAMR), an adaptive framework that jointly controls current-example supervision and historical replay. CAMR combines compatibility checks with a learned memory gate to guide an online policy over three actions: learning incoming examples with general replay, jointly reinforcing compatible related examples, and skipping direct supervision of verified mastered duplicates. The policy learns from acquisition, retention, and training-cost feedback under a bounded per-arrival replay budget. We evaluate CAMR with Qwen3-8B-Base on continual factual updating and cross-task and cross-domain learning. On WikiBigEdit, CAMR consistently leads evaluated methods in Original Match from 3K to 100K facts, reaching 93.66% after just 3K facts and surpassing Full-history Replay (LoRA) after 100K facts. It also achieves the highest final scores on all four selected TRACE tasks and maintains or improves earlier-task performance relative to post-acquisition levels in a sequence spanning financial question answering, mathematical reasoning, code generation, and summarization.

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。

↑