MemComposer jointly examines a task query and a set of retrieved candidate memories, then produces an ordered subset of variable length.
Learned memory assembly helps LLM agents succeed with fewer context tokens
MemComposer selects, sizes and orders retrieved experiences using feedback from complete agent interactions rather than semantic similarity alone.
Big Tech
Yixuan Xie · Xueqiang Xu · Xinyi Fan · Xinxi Lyu · Tao Feng · Xing Fan · +5 more
University of Illinois Urbana-Champaign · Amazon
Research Digest··2 min read
Xie et al.
Why this paper
From Amazon and University of Illinois Urbana-Champaign
In one line
Learning to assemble memories with task success feedback improves LLM agent performance and reduces memory token usage.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§