A learned controller that dynamically selects among valid atomic evidence operations (retrieve, rewrite, gap-query, sufficiency-check, stop) at each step of multi-hop RAG improves F1 on HotpotQA, 2Wiki, and MuSiQue while reducing token consumption.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DynaKRAG: A Unified Framework for Learnable Evidence Control in Multi-Hop Retrieval-Augmented Generation
A learned controller that dynamically selects among valid atomic evidence operations (retrieve, rewrite, gap-query, sufficiency-check, stop) at each step of multi-hop RAG improves F1 on HotpotQA, 2Wiki, and MuSiQue while reducing token consumption.