[論文レビュー] AQR-HNSW: Accelerating Approximate Nearest Neighbor Search via Density-aware Quantization and Multi-stage Re-ranking

Ganap Ashit Tewary, Nrusinga Charan Gantayat+1 more|arXiv (Cornell University)|Feb 25, 2026

Advanced Image and Video Retrieval Techniques被引用数 0

ひとこと要約

AQR-HNSW combines density-aware adaptive quantization, multi-state re-ranking, and SIMD-optimized implementations to speed up HNSW-based ANN search, achieving higher throughput with strong recall and reduced memory usage.

ABSTRACT

Approximate Nearest Neighbor (ANN) search has become fundamental to modern AI infrastructure, powering recommendation systems, search engines, and large language models across industry leaders from Google to OpenAI. Hierarchical Navigable Small World (HNSW) graphs have emerged as the dominant ANN algorithm, widely adopted in production systems due to their superior recall versus latency balance. However, as vector databases scale to billions of embeddings, HNSW faces critical bottlenecks: memory consumption expands, distance computation overhead dominates query latency, and it suffers suboptimal performance on heterogeneous data distributions. This paper presents Adaptive Quantization and Rerank HNSW (AQR-HNSW), a novel framework that synergistically integrates three strategies to enhance HNSW scalability. AQR-HNSW introduces (1) density-aware adaptive quantization, achieving 4x compression while preserving distance relationships; (2) multi-state re-ranking that reduces unnecessary computations by 35%; and (3) quantization-optimized SIMD implementations delivering 16-64 operations per cycle across architectures. Evaluation on standard benchmarks demonstrates 2.5-3.3x higher queries per second (QPS) than state-of-the-art HNSW implementations while maintaining over 98% recall, with 75% memory reduction for the index graph and 5x faster index construction.

研究の動機と目的

• Motivate scalability of HNSW for billion-scale embeddings in production environments.
• Develop a density-aware adaptive quantization scheme that compresses the index without distorting distance relationships.
• Introduce a multi-state re-ranking mechanism to cut unnecessary distance computations.
• Design SIMD-optimized quantization routines to maximize hardware throughput across architectures.

提案手法

• Density-aware adaptive quantization achieving 4x compression while preserving distance relationships.
• Multi-state re-ranking to reduce unnecessary ANN computations by 35%.
• Quantization-optimized SIMD implementations delivering 16-64 operations per cycle across architectures.

実験結果

リサーチクエスチョン

• RQ1 Can density-aware quantization compress HNSW indices by ~4x without harming recall?
• RQ2 Does multi-state re-ranking meaningfully reduce distance computations in ANN search?
• RQ3 How much throughput and memory improvements can be achieved with SIMD-optimized quantization in HNSW-based ANN search?
• RQ4 What are the relative gains in QPS and recall compared to state-of-the-art HNSW on standard benchmarks?

主な発見

• 2.5–3.3x higher queries per second (QPS) than state-of-the-art HNSW implementations while maintaining over 98% recall.
• 75% memory reduction for the index graph.
• 5x faster index construction.