Learning Personalized Memory Gating from Delayed One-Sided Feedback

Key points

• In this paper, we study memory gating at the fact level.1
• We first propose a retrieval-based labeling strategy that labels each fact by whether it is later retrieved and incorporated into responses.1
• 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.1
• The results of our experiment demonstrate that our method outperforms all baselines under different memory budgets and history lengths, establishing it as a promising personalized memory gating policy.1

Sources (1)

• [1]Learning Personalized Memory Gating from Delayed One-Sided Feedback
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 05:34 PM

  Long-term memory enables language-model agents to personalize future interactions, but a finite memory bank cannot preserve every extracted fact.

  In this paper, we study memory gating at the fact level.

Extractive summary: sentences quoted from the sources.