Skip to main content
Search

Maxm Pan

1 related articles

Subscribe to updates
  1. arXiv

    Beyond Outcomes: Dual-View Relational Learning for Efficient Agent Benchmarking

    This work analyzes large-scale execution trajectories from five agent benchmarks, distills six process signals from twelve observable statistics that are systematically associated with final agent performance, and proposes DualViewEval, which jointly models outcome and process relations to learn an exact-size miniset and predict full-benchmark scores, achieving the best results on all five benchmarks, reaching 24×–40× compression on APEX-Agents and BFCL with only 20 tasks, reducing MAE by 14.5%–28.2% over the strongest competitors, and improving Kendall's τ by up to 7.2% relative to EssenceBench on SWE-bench Verified.

You’re all caught up