18 May 2026 in Proceedings of the ACM on Management of Data
Modern retrieval systems increasingly require integrating approximate nearest neighbor search (ANNS) with complex attribute filtering to handle hybrid queries in applications such as recommendation systems and retrieval-augmented generation (RAG). While HNSW-based inline-filtering methods show promise, existing approaches struggle to deliver high throughput under low-selectivity scenarios while balancing search efficiency, filtering generality, and index connectivity. To address these challenges, we propose FAVOR, an efficient filter-agnostic vector ANNS that supports arbitrary filtering conditions while maintaining stable performance across varying selectivity levels. FAVOR introduces three novel features: (1) an integrated architecture that unifies selectivity estimation and filtered ANNS execution, providing a cohesive solution for hybrid vector–attribute queries; (2) a HNSW-based inline-filtering algorithm that introduces an exclusion distance mechanism to dynamically reshape the vector distance distribution, pushing non-target vectors away from the query while promoting valid candidates toward the query, thus improving search efficiency without compromising generality or graph connectivity; and (3) a selectivity-driven search selector that estimates query selectivity and dynamically routes queries between a pre-filtering brute-force algorithm for low-selectivity cases and an optimized HNSW-based search algorithm for other scenarios, ensuring consistent performance. Extensive experiments on real-world datasets demonstrate that FAVOR achieves a 1.3–5× higher QPS at Recall@10 = 95% compared to state-of-the-art methods for arbitrary filtering conditions, while maintaining competitive performance even against tailored solutions in some filtering conditions.
18 May 2026 in Proceedings of the ACM on Management of Data
Learned indexes show promising performance compared with traditional indexes. However, most existing learned index structures rely on fixed heuristics, which limit their robustness under dynamic workloads. In this paper, we present SELIX, a self-designing learned index that automatically generates and optimizes index structures tailored to different workloads. We introduce a unified index template that allows different nodes to adopt structural configurations, node layouts, conflict resolution policies, and search strategies for flexible and fine-grained structural composition. On top of this template, we formulate structural optimization as a learnable function that maps workload states to index configurations, enabling automated search for optimal structures in the vast design space. We train this function using a deep reinforcement learning paradigm enhanced with meta-learning, and propose a dedicated online update strategy, which together enable fast and stable structural adaptation under workload drift. Experimental results show that SELIX consistently outperforms state-of-the-art learned indexes (ALEX, LIPP, FITing-Tree, PGM, XIndex, and FINEdex) in both in-memory and on-disk settings, achieving significantly higher throughput across diverse and dynamic workloads.