Reproduce “Same Trajectory, Different Knowledge: Information-State-Aware On-Policy Distillation for Search Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Minhao Sun, Xian-Ling Mao, Eric Tao, Zi-Ao Ma, Zhijing Wu
ISA-Distill trains search agents to balance using recoverable facts against seeking missing evidence. It estimates recoverability from a student’s no-retrieval conclusions, then uses that estimate to weight teacher feedback favoring fact use or evidence seeking. Across seven question-answering benchmarks, it reaches 48.0% macro-average exact match, versus 47.2% for EviSD; overall search calls decline, while multi-hop calls rise.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.