Traverse organizes long web searches into three states: Rubric, Answer and Verify.
Search agents improve by deciding when to compress and reconsider
Traverse combines answer criteria, independent verification and agent-controlled memory with a reinforcement-learning method designed to stabilize context management.
Chinese Tech
Jingyuan Ma · Lynx Aster · He Zhang · Siyao Song · Weijie Yuan · Zhe Zhang · +2 more
State Key Laboratory of Multimedia Information Processing · Peking University · ByteDance
Research Digest··3 min read
Ma et al.
Why this paper
From ByteDance and 2 others
In one line
A three-state search agent with self-managed memory and final-segment-only reinforcement learning improves long-horizon web search and reaches 72.83 on BrowseComp.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 35B)
- ·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.
§