PROJECTS
CDF-RAG: Causal Dynamic Feedback for Adaptive Retrieval-Augmented Generation
This project introduces CDF-RAG, a novel framework that enhances retrieval-augmented generation (RAG) with causal reasoning. Unlike conventional RAG systems that rely only on semantic similarity, CDF-RAG iteratively refines queries, retrieves structured causal graphs, and performs multi-hop reasoning to distinguish true cause–effect relations from spurious correlations. By validating outputs against causal pathways, the framework improves factual accuracy, causal consistency, and explainability in generative responses. Evaluated across multiple datasets, CDF-RAG demonstrates superior accuracy and reliability, advancing the development of trustworthy, causally grounded LLMs.
Project information
- Category: Future Healthcare, Large Language Models, and Translation and Practice
- Contact Person: Amir M. Rahmani
