AdaRAG dataset

Retrieval-Augmented Generation (RAG) improves the performance of Large Language Models (LLMs) by retrieving and integrating relevant information from external knowledge bases, which helps generate more accurate responses. However, RAG is vulnerable to retrieval corruption attacks, where attackers can induce LLMs to produce inaccurate responses by injecting malicious passages into the retrieval process. In this paper, we propose a novel framework designed to defend against such attacks, called AdaRAG.  We evaluate AdaRAG on four open-domain Question Answering (Q&A) datasets: Natural Questions (NQ), MS-MARCO, HotpotQA, and 2WikiMultiHopQA, using five open-source LLMs. Extensive experiments confirm the effectiveness and generalization of the AdaRAG pipeline against retrieval corruption attacks.