AI TrackerFrontier desk · AEST
Back to news
normalaitech

CapCover Filter: Semantic Approximate Membership Queries over Embeddings

arXiv cs.AI · 16 Nov 2026, 03:01 AEST

The CapCover Filter enhances query throughput in vector-store systems by enabling Semantic Approximate Membership Queries (SAMQs). It significantly reduces index accesses while maintaining high performance across various datasets.

Key facts

Read aloud

The CapCover Filter enhances query throughput in vector-store systems by enabling Semantic Approximate Membership Queries SAMQs. It significantly reduces index accesses while maintaining high performance across various datasets. Determining whether an embedding is sufficiently similar to any vector in a vector store without performing a nearest-neighbor search can substantially improve query throughput in vector-store systems. We introduce CapCover Filter, a geometric data structure that summarizes a set of embeddings using spherical caps. Across five real-world datasets, at a cosine similarity threshold of zero point nine five, CapCover Filter eliminates approximately twenty--one hundred percent of index accesses while using only two percent of the raw FP32 embedding memory. When placed before a FAISS flat index, it improves query throughput by up to three hundred eighty two percent and by approximately one hundred sixteen percent on average across the five datasets, while returning the same range-search result.

Sources in this cluster