arXiv cs.AI· BHVSP Subrahmanyam, Neeraj Choudhary·· 12 小时前AI 评分31
超越边际影响：面向 LLM 智能体的交互感知记忆选择
Beyond Marginal Influence: Interaction-Aware Memory Selection for LLM Agents
AI 导读
研究提出 JointCoreBench 基准与 JointCore 基线，用于 LLM 智能体的逐决策记忆充分性选择。JointCoreBench 覆盖九类交互族、六个工作流领域，在 6,912 个精确条件下，成本排序的 JointCore 100.0% 恢复最小成本充分集，子集评估从 30.00 降至 11.13（62.9%），节省 87.8% 活跃记忆。
正文 · 原文
Abstract:LLM-agent memory selection often uses retrieval rankings based on record relevance or individual influence. Such rankings cannot identify sufficient subsets under redundant or jointly necessary evidence. We study per-decision memory sufficiency: selecting a small subset that preserves full-memory behavior. To separate search error from model error, we introduce JointCoreBench, a benchmark with nine interaction families, six workflow domains, natural-language decisions, and exact sufficient-set oracles, and JointCore, a relevance cost-ordered group-deletion baseline that tests blocks before individual records. On 6,912 exact conditions, cost-ordered JointCore recovers a minimum-cost sufficient set in 100.0% of cases, reduces subset evaluations from 30.00 to 11.13 (62.9%), and achieves 87.8% active-memory savings. Across four 12B-70B model families and 432 tasks per model, its unweighted mean is 10.28 versus 16.00 model queries and 50.6% versus 43.1% minimum-cost recovery relative to cost aware singleton deletion. Both methods preserve full-memory behavior under the deterministic primary interface. Tests with physical deletion, UNKNOWN masks, PENDING sentinels, counterfactual values, and alternative labels identify interface dependent behavior. JointCore is an efficient benchmark baseline; the study does not establish deployment-ready memory compression.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.23266 [cs.LG] |
| (or arXiv:2610.23266v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23266 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Neeraj Choudhary [view email]
[v1]
Fri, 16 Oct 2026 10:37:47 UTC (67 KB)
来源：arXiv cs.AI · arxiv.org