, after context compaction or subagent forking).
New RL method trains AI agents to handle multiple sessions effectively
Mosaic uses hierarchical advantage estimation, outperforming broadcasting by up to 10.1% accuracy while using fewer sessions
Chinese Tech
Yilong Xu · Minrui Xu · Xinyu Zhou · Xiang Long · Haochen Tan · Zhijiang Guo
HKUST (Guangzhou) · Huawei · City University of Hong Kong · HKUST
Research Digest··2 min read
Xu et al.
Why this paper
From Huawei and 3 others
In one line
Mosaic uses hierarchical advantage optimization to outperform broadcasting in multi-session agentic RL on BrowseComp by 10.1% and with fewer sessions.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§