Know What You Don't Know: Agentic Data Synthesis by Evolving Generation Space for Closed-Loop Learning
3.40T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.12593.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryEvoSyn is a learning-feedback-driven multi-agent framework that addresses the feedback-to-decision gap in synthetic data generation for training interactive agents. A Designer-Planner role-specialized system evolves the synthesis space based on learner weaknesses. It improves average task success by 13.3 points over the strongest synthesis baseline on τ²-Bench and CAR-Bench.
Why it mattersConcrete benchmark gains and an explicit framing of how learner failures can drive revisions to the training data space. Useful reference for anyone building self-improving agent training loops.
Cited by
No citations on record.