arXiv cs.AI· Bingfan Zeng, Zhisheng Chen, Chenbo Sang, Zhengwei Xie, Jinpeng Wang, Xiangchen Guan, Rui Qian, Zheng Lu, Jingwei Song·· 9 天前AI 评分27
MemoWM：世界模型如何改变智能体的记忆需求
MemoWM: How World Models Change What Agents Need to Remember
AI 导读
MemoWM 框架利用世界模型的共享预测压缩智能体保留的信息并重建被省略内容，在五个长期智能体记忆基准上平均答案准确率达 42.42%，超出最强基线 2.62 个百分点，同时相比存储效率最高的基线 MIRIX 将平均单条经验存储降低 53.9%。分析显示，更强的世界模型可在任务质量相当的情况下减少每条经验的存储，而模型参数与存储成本之间存在权衡，保留的交互越多，最小化总存储所需容量越大。
正文 · 原文
Abstract:Long-term agents face growing storage demands as they accumulate experience. World models capture reusable regularities that can reduce the information stored for each experience. We formulate the problem of memory allocation conditioned on a world model and introduce MemoWM, a framework that uses shared predictions to compress retained information and reconstruct omitted content. Its task-aware allocation rule balances the expected impact of reconstruction errors against storage cost, retaining information with downstream value beyond the predictive prior. Across five long-term agent-memory benchmarks, MemoWM achieves 42.42\% average answer accuracy, exceeding the strongest baseline by 2.62 percentage points, while reducing average experience-specific storage by 53.9\% relative to MIRIX, the most storage-efficient baseline. Further analysis shows that stronger world models reduce per-experience storage at comparable task quality. Accounting for model parameters reveals a trade-off between shared model capacity and recurring storage costs, with the capacity that minimizes total storage increasing as more interactions are retained. Our code is available at this https URL.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.10778 [cs.LG] |
| (or arXiv:2610.10778v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.10778 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zhisheng Chen [view email]
[v1]
Wed, 7 Oct 2026 18:33:00 UTC (946 KB)
来源：arXiv cs.AI · arxiv.org