uva-irlab-conv at SemEval-2026 Task 8: Multi-Turn RAG with Learned Sparse Retrieval and Listwise Reranking
Researchers have developed ProGRank, a novel defense mechanism designed to protect Retrieval-Augmented Generation (RAG) systems from corpus poisoning attacks. This training-free method operates on the retriever side by introducing mild perturbations to query-passage pairs and analyzing probe gradients to identify instability signals. Separately, another research team details their participation in SemEval-2026 Task 8, presenting a multi-turn RAG pipeline that integrates learned sparse retrieval with LLM-based reranking for improved conversational question answering across various domains. AI
IMPACT These papers introduce novel techniques for enhancing RAG security and improving multi-turn conversational AI performance, potentially impacting future development in both areas.