Rethinking Memory in LLM based Agents: Representations, Operations, and Emerging Topics
YIMING DU, WENYU HUANG, DANNA ZHENG, ZHAOWEI WANG, Sebastien Montella, Mirella Lapata, Kam-Fai Wong, Jeff Z. Pan · 2025
Read Rethinking Memory in LLM based Agents: Representations, Operations, and Emerging Topics by YIMING DU, WENYU HUANG, DANNA ZHENG, ZHAOWEI WANG, Sebastien Montella, Mirella Lapata, Kam-Fai Wong, Jeff Z. Pan (2025) 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-2505-00675.txt.
Description
Yiming Du and co-authors rethink how memory should be studied in language model agents, in a 2025 survey. Memory is basic to such agents, yet past surveys stress uses such as personal dialogue and skip the small operations that drive memory change. The authors split memory into two forms: one hidden in model weights, and one held as outside data.
Six core operations are defined: consolidation, updating, indexing, forgetting, retrieval, and condensation. Crossing those operations with the two forms brings out four key research topics, covering long-term memory, long context, changing the weights, and memory from many sources. The taxonomy maps papers, benchmarks, and tools in a structured way. Data sets and tools are shared in public. The authors aim to guide future work on agent memory.
Catalog record
| CREATOR | YIMING DU, WENYU HUANG, DANNA ZHENG, ZHAOWEI WANG, Sebastien Montella, Mirella Lapata, Kam-Fai Wong, Jeff Z. Pan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 159 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-00675