agent/papers/items/2026-2606-25161-trustmem-learning-trustworthy-memory-consolidation-for-llm-agents-with-long-term.md

Paper: TRUSTMEM: Learning Trustworthy Memory Consolidation for LLM Agents with Long-Term Memory

---

type: paper title: "TRUSTMEM: Learning Trustworthy Memory Consolidation for LLM Agents with Long-Term Memory" authors: Tianyu Yang, Sudipta Paul, Vijay Srinivasan, Vivek Kulkarni, Srinivas Chappidi year: 2026 venue: arXiv url: https://arxiv.org/abs/2606.25161 code_url: source: arxiv collected_at: 2026-07-08 published_at: 2026-06-23 updated_at: 2026-06-23 status: skimmed relevance: high topics:

• agent-evaluation
• computer-use
• memory
• rag
• reasoning
• tool-use methods:
• memory-transition-verification
• transition-ranked-grpo
• write-revise-prune benchmarks:
• MemoryAgentBench
• HaluMem
• Mem-alpha models:

datasets:

• cs.AI related_concepts:
• memory-consolidation
• transition-verification related_jobs:

related_experiments:

• KC-001-agent-memory-pilot related_projects:
• learning/agent-memory collection_score: 19 collection_queries: agent-memory

---

One-line Takeaway

持久 memory 需要在每次 WRITE、REVISE、PRUNE 后检查 coverage、preservation 和 faithfulness，不能只看最终回答是否正确。

Pilot Skim

• problem: 终局 reward 无法定位 memory 更新中的遗漏、破坏和幻觉，而错误会成为持续系统状态。
• method: Memory Transition Verifier 为局部更新打分，并用 Transition-Ranked GRPO 优化更新策略。
• evidence: 相比各错误类型最强 baseline，遗漏、破坏、幻觉分别减少 40.1%、79.1%、50.0%。
• boundary: verifier 本身仍由冻结 LLM 实现；内容可靠不等于来源可信或具备行动权限。

Used In

• Agent Memory evidence matrix

Links

• arXiv: https://arxiv.org/abs/2606.25161