HaS: Accelerating RAG through Homology-Aware Speculative Retrieval

A research team led by Peng Peng has introduced HaS (Homology-Aware Speculative Retrieval), a novel framework designed to tackle the growing latency problem in Retrieval-Augmented Generation (RAG) systems. As knowledge databases expand, retrieving external documents for LLM context becomes a major bottleneck. Existing solutions either sacrifice accuracy with approximations or offer minimal speed gains by caching only identical queries. HaS innovates by speculatively retrieving documents from a restricted scope and then validating them based on a 'homology relation' between the new query and previously seen ones. This validation is framed as a homologous query re-identification task, allowing the system to bypass the slow, full-database search when a match is found.

Extensive experiments demonstrate HaS's practical efficiency. The framework achieves latency reductions of 23.74% and 36.99% across different datasets while incurring a marginal accuracy drop of just 1-2%. This performance boost stems from the real-world prevalence of similar queries under common popularity patterns. Crucially, HaS is designed as a plug-and-play module, meaning it can be integrated into existing RAG and modern agentic AI pipelines—which often involve complex, multi-hop reasoning—to deliver significant speed improvements without a complete system overhaul. The paper has been accepted for presentation at the ICDE 2026 conference, and the source code is publicly available, paving the way for broader adoption and testing.