Back to AI News
arxiv

Personalized Memory Rubrics for Per-User Credit Assignment in Long-Horizon Agents

Researchers introduce Personalized Memory Rubrics (PMR), a training framework that converts terminal question-answering feedback into fine-grained, per-step and per-user credit for memory updates in long-horizon personalized agents. PMR splits evaluation into a universal layer for cross-persona failure patterns and an adaptive per-user layer, then performs token-level rubric-guided credit assignment that amplifies validated decisions on successful trajectories and falls back to standard policy optimization on failed ones. The rubric and memory policy co-evolve in a closed loop with anti-circularity guarantees, and the method adds no inference overhead or external judge at deployment. Experiments across four benchmarks report consistent gains over strong memory-augmented baselines.

AI Brief

It offers a concrete training method for making personalized agent memory update policies user-specific without adding inference-time cost.

arXiv:2611.14017v1 Announce Type: new Abstract: Long-horizon personalized agents maintain an evolving memory by continuously deciding what to add, modify, or discard. However, learning an optimal update policy remains challenging. Conventional training relies solely on terminal question-answering accuracy, yielding a coarse feedback signal that treats all intermediate updates identically. Furthermore, update utility is inherently user-specific, as identical memory actions may yield opposite effects across different user personas. Existing methods fail to address this heterogeneity, relying on globally shared evaluation standards with at most a single scalar intermediate reward. To address these challenges, we propose \textbf{Personalized Memory Rubrics} (\textbf{PMR}), a framework that translates terminal feedback into fine-grained, per-step, and per-user credit assignment. PMR decomposes evaluation into a universal layer capturing cross-persona failure patterns and a per-user adaptive layer modeling individual traits as a specialized delta over the universal baseline. Through \textbf{Rubric-Guided Credit Assignment}, PMR redistributes credit at the token level, selectively amplifying validated decisions on successful trajectories while falling back to standard policy optimization on failed ones to avoid unreliable attribution. The rubric and memory policy \textbf{co-evolve} in a closed loop with anti-circularity guarantees. Crucially, PMR incurs zero inference overhead, operating without any rubric or external judge during deployment. Extensive experiments across four benchmarks demonstrate consistent gains over strong memory-augmented baselines.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!