analiza aiqbench
CoEM addresses the degradation of large language models in long-context scenarios by introducing a dynamic memory management system that delays irreversible compression until evidence is verified. By leveraging reinforcement learning to decide when to commit source excerpts to memory, this approach significantly improves reasoning accuracy on tasks involving thousands of documents without sacrificing critical details. The method represents a notable advancement in handling extended context windows for complex, multi-step AI tasks.
CoEM: Empowering Long-Context Reasoning with Commit-on-Evidence Memory
arXiv:2609.36935v1 Announce Type: new Abstract: Long-context reasoning is essential for complex and long-horizon tasks, yet the performance of large language models (LLMs) degrades as context length increases. Recent approaches address this by processing input chunk by chunk while maintaining a bounded textual memory in model context. However, premature information compression can discard critical details essential for subsequent reasoning. In this paper, we introduce Commit-on-Evidence Memory (CoEM), which learns when to convert source evidence into compact memory facts. Specifically, under a fixed context-memory budget, CoEM preserves potentially useful source excerpts verbatim in a pending set, allowing subsequent context to clarify their relevance before irreversible compression. As new context arrives, a learned policy revisits each pending excerpt and decides whether to promote it to the committed memory, retain it for further consideration, or discard it. A frozen verifier ensures proposed facts are accepted only if supported by retained excerpts and current context. To further guide effective memory management, we train this policy using reinforcement learning by combining fine-grained, step-level evidence rewards with final answer rewards. Extensive experiments demonstrate that CoEM consistently improves long-context reasoning. When evaluated on 6,400 documents long-context input, CoEM outperforms the strongest memory baseline by 10.4-11.4 F1 points on Qwen3.5-9B. Code repository: https://github.com/benmagnifico/CoEM.
Abstract:Long-context reasoning is essential for complex and long-horizon tasks, yet the performance of large language models (LLMs) degrades as context length increases. Recent approaches address this by processing input chunk by chunk while maintaining a bounded textual memory in model context. However, premature information compression can discard critical details essential for subsequent reasoning. In this paper, we introduce Commit-on-Evidence Memory (CoEM), which learns when to convert source evidence into compact memory facts. Specifically, under a fixed context-memory budget, CoEM preserves potentially useful source excerpts verbatim in a pending set, allowing subsequent context to clarify their relevance before irreversible compression. As new context arrives, a learned policy revisits each pending excerpt and decides whether to promote it to the committed memory, retain it for further consideration, or discard it. A frozen verifier ensures proposed facts are accepted only if supported by retained excerpts and current context. To further guide effective memory management, we train this policy using reinforcement learning by combining fine-grained, step-level evidence rewards with final answer rewards. Extensive experiments demonstrate that CoEM consistently improves long-context reasoning. When evaluated on 6,400 documents long-context input, CoEM outperforms the strongest memory baseline by 10.4-11.4 F1 points on Qwen3.5-9B. Code repository: this https URL.
38 pages, 13 figures. Code repository: this https URL
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2609.36935 [cs.AI]
(or arXiv:2609.36935v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.36935 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 308 cuvinte
Subiecte conexe
Deconfusing Corrigibility
AI Alignment Forum · 17 oct. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
VISOR: Agentic Visual Retrieval-Augmented Generation via Iterative Search and Over-horizon Reasoning
arXiv · 17 aug. 2026
How GoFundMe became America’s backup plan
The Verge · 24 aug. 2026