统一无任务持续学习中的检测与适配
Unifying Detection and Adaptation in Task-Free Continual Learning
浏览论文内容
中文总结 AI 辅助
本文提出FiUni框架，通过K-FAC近似的Fisher主子空间正交性检测批量级任务，结合LoRA实现参数高效适配，动态平衡知识共享与任务隔离，缓解LLM持续学习的灾难性遗忘，性能优于先进方法且参数更少。
中文摘要 AI 辅助
为缓解大语言模型（LLM）下游持续学习（CL）中的灾难性遗忘问题，现有方法通常会限制参数更新或引入任务特定适配模块，但这些方法往往依赖训练时的显式任务边界，限制了其在现实无任务场景中的适用性。本文提出一种用于批量级任务检测与参数高效持续适配的Fisher引导统一（FiUni）框架。FiUni的核心动机源于对预训练模型Fisher信息矩阵（FIM）的关键观察：从少量下游任务样本估计的Kronecker分解近似曲率（K-FAC）近似的主子空间之间的正交性，可反映不同任务间的相似性。基于此观察，FiUni构建FIM衍生的冻结子空间以引导低秩适配（LoRA），同时将每个传入批量窗口的Fisher主子空间与历史子空间匹配，从而自适应决定是复用现有知识、扩展相关子空间还是创建新子空间，动态平衡知识共享与任务隔离。实验表明，FiUni可有效推断潜在批量级任务归属，且在可训练参数更少的情况下，取得与先进任务感知CL方法相当的性能。
英文摘要
To mitigate catastrophic forgetting in downstream continual learning (CL) for large language models (LLMs), existing methods typically constrain parameter updates or introduce task-specific adaptation modules. However, these methods often rely on explicit task boundaries during training, limiting their applicability to realistic task-free scenarios. In this paper, we propose a \textbf{Fi}sher-guided \textbf{uni}fied (\textbf{FiUni}) framework for batch-level task detection and parameter-efficient continual adaptation. FiUni is motivated by a key observation about the Fisher information matrix (FIM) of pre-trained models: the orthogonality among the principal subspaces of its Kronecker-Factored Approximate Curvature (K-FAC) approximation, estimated from a small number of downstream task samples, can reflect the similarity between different tasks. Based on this observation, FiUni constructs FIM-derived frozen subspaces to guide low-rank adaptation (LoRA), while matching the Fisher principal subspace of each incoming batch window with historical subspaces. This enables FiUni to adaptively determine whether to reuse existing knowledge, expand a related subspace, or create a new subspace, dynamically balancing knowledge sharing and task isolation. Experiments show that FiUni can effectively infer latent batch-level task affiliations and achieve competitive performance against advanced task-aware CL methods with fewer trainable parameters.
发表机构
- School of Control Science and Engineering, Shandong University（山东大学控制科学与工程学院）
- The Hong Kong University of Science and Technology（香港科技大学）
- Shandong University（山东大学）
- Qilu Hospital of Shandong University（山东大学齐鲁医院）
机构由 AI 辅助整理，请以论文原文为准。
