UMEM: Unified Memory Extraction and Management Framework for Generalizable Memory
Yongshi Ye, Hui Jiang, Feihu Jiang, Tian Lan, Yichao Du, Biao Fu, Xiaodong Shi, Qianghuai Jia, Longyue Wang, Weihua Luo · 2026
Read UMEM: Unified Memory Extraction and Management Framework for Generalizable Memory by Yongshi Ye, Hui Jiang, Feihu Jiang, Tian Lan, Yichao Du, Biao Fu, Xiaodong Shi, Qianghuai Jia, Longyue Wang, Weihua Luo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-10652.txt.
Description
Yongshi Ye and nine co-authors unify how AI agents build and maintain memory in a 2026 paper. Self-evolving memory acts as trainable settings for language-model agents, but current methods optimize memory management while treating extraction as static. Agents then accumulate instance-exact noise instead of robust memories that transfer.
Their system, UMEM, jointly optimizes a language model to extract and manage memories jointly. Meaning neighborhood modeling plus a neighborhood-level marginal utility reward counter overfitting to single instances. By judging a memory's utility across its meaning neighborhood, the system keeps memories that transfer and discards ones that merely fit one episode. The paper reports improved transfer for agents that must learn continuously.
Catalog record
| CREATOR | Yongshi Ye, Hui Jiang, Feihu Jiang, Tian Lan, Yichao Du, Biao Fu, Xiaodong Shi, Qianghuai Jia, Longyue Wang, Weihua Luo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10652