CoRE learns rule routing and constrained retrieval without labeled pairs
Original title: Co-adaptive Rule Routing and Constraining for Structured Knowledge Retrieval
A router and rule-constrained retriever co-adapt with reinforcement and contrastive learning on STaRK-QA.
CoRE jointly learns which reasoning rule to assign to a query and how to retrieve evidence under that rule, without predefined query-rule pairs. A routing agent selects a rule; a retriever traverses the knowledge base and ranks evidence that meets the rule’s constraints. Training alternates reinforcement learning for the router with contrastive learning for the retriever. Experiments cover scientific discovery, biomedical reasoning, and e-commerce on STaRK-QA, and the authors say generation-based routing transfers across datasets.
- No predefined query-rule supervision is required.
- Routing uses reinforcement learning; retrieval uses contrastive learning.
- Selected rules supervise the retriever, whose performance rewards routing.
- Generation-based routing benefits from multi-domain pretraining then fine-tuning.
- AI models
- CoRE
From the original article
1 min read at huggingface.co
Structured knowledge retrieval acquires task-relevant evidence from structured knowledge sources for downstream tasks. Existing methods rely on unconstrained retrieval or predefined query-rule supervision, such as author-paper rules for scientific discovery and anatomy-drug rules for precision medicine. Such supervision is rarely available in practice.
By Zhisheng Qi, Haoyu Han, Jiaxi Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.