[论文] Learning Personalized Memory Gating from Delayed One-Sided Feedback

论文概要

研究领域: ML 作者: Kaixuan Ji, Zixiang Chen, Wenting Zhao, Cheng Qian, Heng Wang, Weizhi Zhang, Jielin Qiu, Liangwei Yang, Quanquan Gu, Silvio Savarese, Shelby Heinecke, Huan Wang 发布时间: 2026-10-16 arXiv: 2610.23812

中文摘要

长期记忆使语言模型智能体能够个性化未来的交互，但有限的记忆库无法保留每一条提取的事实。现有的记忆门控策略通常在用户间共享，无法考虑用户特定的未来复用模式。本文在事实层面研究记忆门控。我们首先提出一种基于检索的标注策略，根据每条事实是否被后续检索并融入回应来标注。随后将在线记忆门控建模为具有延迟单侧反馈的情境苹果品尝问题，并提出 Warm-Start UCB——结合离线学习（从现有数据）和在线适应（针对当前用户）。我们的理论分析建立了高概率遗憾上界 Õ(Tε∧d√T)（延迟窗口至多为 O(√T)），其中 T 是总轮数，ε 是初始化误差，d 是特征映射维度，Õ(·) 隐藏常数和多对数因子。实验结果表明，我们的方法在不同记忆预算和历史长度下均优于所有基线，确立了其作为有前景的个性化记忆门控策略的地位。

原文摘要

Long-term memory enables language-model agents to personalize future interactions, but a finite memory bank cannot preserve every extracted fact. Existing memory gating policies are typically shared across users and therefore fail to account for user-specific patterns of future reuse. In this paper, we study memory gating at the fact level. We first propose a retrieval-based labeling strategy that labels each fact by whether it is later retrieved and incorporated into responses. We then formulate online memory gating as contextual apple tasting with delayed one-sided feedback, and propose Warm-Start UCB, which combines offline learning from existing data with online adaptation to the current user. Our theoretical analysis establishes a high-probability regret upper bound of $\tilde{O}(Tε\w...

#论文 #arXiv #ML #小凯