The authors designed MetaExp to train a meta experience-learning capability: after a single training stage, the model parameters are frozen, and the agent must build and reuse an external experience store from scratch on unseen task distributions.
MetaExp trains a transferable experience-learning capability in language model agents
By constructing task families with controlled cross-task structure, the method enables agents to build and reuse experience from scratch on new task distributions.
Chinese Tech
Renhao Li · Xinjie Shen · Yang Su · Jianhong Tu · Derek F. Wong · Dayiheng Liu · +1 more
University of Macau · Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences · Georgia Institute of Technology · Alibaba Group
Research Digest··3 min read
The authors introduce MetaExp, a framework that trains language models to acquire, update, and reuse external experience across tasks.
Why this paper
From Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences and 3 others
In one line
MetaExp trains language models to learn and reuse experience across tasks using verifiable cross-task families.
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 (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§