Fairness-Aware Retrieval Optimization for Retrieval-Augmented Generation
Two new research papers explore methods to improve the reliability and fairness of Retrieval-Augmented Generation (RAG) systems. One paper introduces BiRD, a defense mechanism that uses bidirectional ranking to detect and mitigate adversarial poisoning attacks, significantly reducing attack success rates while maintaining task accuracy. The other paper proposes a fairness-aware retrieval framework that models and controls bias introduced during the retrieval process, aiming to balance relevance and fairness in RAG outputs. AI
IMPACT New research offers methods to enhance RAG system security against attacks and improve fairness, potentially increasing trust and adoption.