analiza aiqbench
This research introduces Unified Memory Agent (UMA), a framework that enables proactive, learned memory management for long-context reasoning by consolidating state tracking outside of query time. By utilizing a task-stratified reinforcement learning approach, the system maintains a reusable external memory bank that significantly outperforms standard retrieval-augmented methods on evolving fact scenarios. The work demonstrates that decoupling evidence consolidation from immediate queries creates more robust agents capable of handling long-horizon tasks without brittle state tracking.
Learning to Remember: End-to-End Training of Memory Agents for Long-Context Reasoning
arXiv:2602.18493v2 Announce Type: replace Abstract: Long-context LLMs and Retrieval-Augmented Generation defer state tracking and evidence consolidation to query time, which is brittle when facts evolve and answers depend on latent states. We introduce Unified Memory Agent (UMA) for a one-to-many setting: query-agnostic external memory is constructed once from a stream and reused across multiple future QA sessions. A single policy maintains a structured Memory Bank through CRUD operations and answers using both the Memory Bank and raw context. Task-Stratified GRPO uses the mean reward of QA trajectories branching from each sampled memory state to supervise memory maintenance, while normalizing memory and per-question QA groups separately. We also introduce Ledger-QA, a diagnostic benchmark for long-horizon state tracking over accumulated updates. At the 16k budget, UMA-Generalist achieves the highest average score among compared methods across the test-time-learning and accurate-retrieval benchmarks and transfers to Ledger-QA without task-specific training; UMA-Specialist further improves long-horizon tracking after task adaptation. These results support learned proactive memory management for long-context reasoning.
Abstract:Long-context LLMs and Retrieval-Augmented Generation defer state tracking and evidence consolidation to query time, which is brittle when facts evolve and answers depend on latent states. We introduce Unified Memory Agent (UMA) for a one-to-many setting: query-agnostic external memory is constructed once from a stream and reused across multiple future QA sessions. A single policy maintains a structured Memory Bank through CRUD operations and answers using both the Memory Bank and raw context. Task-Stratified GRPO uses the mean reward of QA trajectories branching from each sampled memory state to supervise memory maintenance, while normalizing memory and per-question QA groups separately. We also introduce Ledger-QA, a diagnostic benchmark for long-horizon state tracking over accumulated updates. At the 16k budget, UMA-Generalist achieves the highest average score among compared methods across the test-time-learning and accurate-retrieval benchmarks and transfers to Ledger-QA without task-specific training; UMA-Specialist further improves long-horizon tracking after task adaptation. These results support learned proactive memory management for long-context reasoning.
Accepted at EMNLP 2026 Main Conference. Camera-ready version
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2602.18493 [cs.LG]
(or arXiv:2602.18493v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2602.18493 Focus to learn more arXiv-issued DOI via DataCite
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 · 261 cuvinte
Subiecte conexe
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · 12 iun. 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · 25 sept. 2026
Biological Superintelligence
LessWrong · 30 iul. 2026
Academia is for Ambition — Alex Zhang, MIT
Latent Space · 2 oct. 2026