TRIO: Task Refinement with Individual Optimization for Retrieval-Augmented Generation
TRIO trains three agents separately to sharpen queries, judge retrieval sufficiency and improve answers.
The query agent expands ambiguous requests and receives rewards tied to improved answer quality. Unanimity voting supplies sufficiency labels used to train the judge agent. The answer agent learns to use evidence through dynamic matching rewards and curriculum learning. The authors report gains over state-of-the-art approaches across multiple benchmarks, but the abstract gives no numerical results.
HF Daily Papers' note
The query agent expands ambiguous requests and receives rewards tied to improved answer quality. Unanimity voting supplies sufficiency labels used to train the judge agent. The answer agent learns to use evidence through dynamic matching rewards and curriculum learning. The authors report gains over state-of-the-art approaches across multiple benchmarks, but the abstract gives no numerical results.
HF Daily Papers' note
score 4