The authors address credit assignment in long-horizon deep search agents trained with reinforcement learning.
KnowGPO aligns trajectories by knowledge for better credit assignment in search agents
The method groups search steps by the query-specific knowledge they have acquired and uses incremental knowledge gain as a process reward, outperforming outcome-only and observation-based approaches.
Chinese Tech
Siqi Chen · Zhengxi Lu · Yuchen Yan · Xingyu Wu · Qipeng Chen · Xin Zhang · +7 more
Zhejiang University · Tencent
Research Digest··3 min read
The authors introduce KnowGPO, a reinforcement learning framework that assigns credit in deep search agents by representing each step's state as a subset of atomic knowledge conditions satisfied, rather than by surface observations.
Why this paper
From Tencent and Zhejiang University
In one line
KnowGPO aligns trajectories by knowledge state, outperforming baselines by 3.8-4.2 points on deep search benchmarks.
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.
§