arxiv: 2605.07770 · v1 · submitted 2026-05-08 · 💻 cs.IR

Recognition: no theorem link

FAVOR: Efficient Filter-Agnostic Vector ANNS Based on Selectivity-Aware Exclusion Distances

Beng Chin Ooi, Guoyu Hu, Junjie Song, Ke Zhou, Ming Yang, Yu Liu, Zhongle Xie

Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:34 UTC · model grok-4.3

classification 💻 cs.IR

keywords approximate nearest neighbor searchvector ANNSattribute filteringHNSWselectivity estimationhybrid queriesinline filteringexclusion distance

0 comments

The pith

FAVOR uses exclusion distances in HNSW graphs to push non-matching vectors away, delivering faster filtered vector search across all selectivity levels.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces FAVOR to solve the problem of running approximate nearest neighbor search when queries also include arbitrary attribute filters. It combines a new exclusion distance that reshapes the distance distribution inside the graph, a selectivity estimator that decides whether to use brute-force pre-filtering or graph search, and a single integrated index structure. The goal is to keep high recall while raising queries per second for hybrid workloads that current inline or pre-filter methods handle poorly when selectivity varies. If the approach works, retrieval systems could support complex recommendation and generation queries without maintaining separate indexes or suffering slowdowns on low-selectivity filters.

Core claim

FAVOR presents a filter-agnostic ANNS architecture built on HNSW that adds an exclusion distance mechanism to dynamically increase distances to non-target vectors and decrease distances to valid candidates, thereby improving search efficiency without altering graph connectivity. A selectivity-driven selector estimates the fraction of vectors satisfying the filter and routes low-selectivity queries to a pre-filtering brute-force path while directing the rest to the optimized inline HNSW algorithm. Experiments on real-world datasets show this yields 1.3-5× higher QPS at Recall@10 = 95% versus prior methods for arbitrary filters.

What carries the argument

The exclusion distance mechanism inside HNSW-based inline filtering, which reshapes the vector distance distribution to favor target vectors for a given filter condition.

If this is right

Consistent high throughput is obtained for arbitrary filter conditions instead of requiring filter-specific index variants.
Low-selectivity queries avoid the inefficiency of traversing the full graph by switching to pre-filtering.
Graph connectivity remains intact, so the same index supports both unfiltered and filtered searches.
Selectivity estimation is integrated into the query path rather than treated as a separate preprocessing step.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same exclusion-distance idea could be tested on other proximity graphs such as NSG or Vamana to see whether the gains transfer.
Production systems might reduce the number of separate filtered indexes they maintain if one FAVOR-style index covers the full range of selectivity.
The approach opens a route to adaptive query planning that also considers index size or memory constraints when choosing the search path.

Load-bearing premise

That the exclusion distance adjustment improves search speed without breaking the connectivity or reachability of valid candidates in the graph.

What would settle it

Measure recall and QPS on a dataset where many vectors have nearly identical attributes; if recall at 95% drops below the baseline or QPS gains disappear, the exclusion distance is not preserving search quality.

Figures

Figures reproduced from arXiv: 2605.07770 by Beng Chin Ooi, Guoyu Hu, Junjie Song, Ke Zhou, Ming Yang, Yu Liu, Zhongle Xie.

**Figure 1.** Figure 1: Workflow of FAVOR. 3.1 Key idea We aim to develop an efficient filtered ANNS method that supports arbitrary F while maintaining high QPS across the variant selectivity range. 3.1.1 Key Idea. Our key idea is that high performance for filtered ANNS within the HNSW framework can be achieved by increasing the number of TD points that are closer to 𝒒 than their neighbors on the search path. Existing approaches … view at source ↗

**Figure 2.** Figure 2: FAVOR search process. ensures that PreFBF is chosen when selectivity falls below 𝜆, while the HNSW-based algorithm is used otherwise. We determine the 𝜆 = 1% based on empirical measurement. We refer to the interval in which 𝑝 is close to but slightly exceeds 1% as the middle selectivity range. Since it is a less representative scenario [8, 11], we adopt a conservative retrieval strategy that prioritizes gr… view at source ↗

**Figure 3.** Figure 3: The vector distance distributions under various [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

**Figure 4.** Figure 4: Search performance in Equality_bool ( (a) – (g) ) and Equality_int ( (h) – (n) ). 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 QPS (a)SIFT1M 0.8 0.9 1.0 Recall@10 10 1 10 2 10 3 (b)GIST1M 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (c)DEEP1M 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (d)Msong 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (e)Paper 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (f)Words 0.8 0.9 1.0 Recall@10 10 1 10… view at source ↗

**Figure 6.** Figure 6: Search performance at different recall levels. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗

**Figure 7.** Figure 7: QPS yielded by different methods across varying selectivity at [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗

**Figure 8.** Figure 8: Search performance in Range_10 ( (a) – (g) ) and Range_50 ( (h) – (n) ). ACORN-1 ACORN- Result Set Filtering Milvus Vearch FAVOR 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 QPS (a)SIFT1M 0.8 0.9 1.0 Recall@10 10 1 10 2 10 3 (b)GIST1M 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (c)DEEP1M 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (d)Msong 0.90 0.95 1.00 Recall@10 10 1 10 2 10 3 (e)Paper 0.90 0.95 1.00 Recall@10 10 1 1… view at source ↗

**Figure 9.** Figure 9: Search performance in Logic. 0.8 0.9 1.0 Recall@10 0.5 1.0 QPS 1e3 SIFT1M 0.70 0.85 1.00 Recall@10 0 1 2 1e2 GIST1M 0.8 0.9 1.0 Recall@10 0.5 1.0 1.5 1e3DEEP1M D0 +Dmax +D [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗

**Figure 10.** Figure 10: Effect of exclusion distance strategies. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗

**Figure 11.** Figure 11: Impact of optimized termination condition. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗

**Figure 12.** Figure 12: Performance vs. TD proportion on search paths. [PITH_FULL_IMAGE:figures/full_fig_p022_12.png] view at source ↗

**Figure 13.** Figure 13: Comparison of search path length and QPS. [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗

read the original abstract

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$\times$ 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

FAVOR adds exclusion-distance reshaping to HNSW plus a selectivity router, but offers no analysis of how the distance changes affect traversal paths or filtered recall.

read the letter

The one thing to take away is that FAVOR tries to solve the low-selectivity problem in filtered ANNS by dynamically increasing distances to non-matching vectors inside an existing HNSW graph and then routing low-selectivity queries to brute force. The index itself stays untouched, which is a practical choice. The selectivity estimator and the dynamic switch between paths are the parts that feel new relative to earlier inline-filtering HNSW papers. The authors show the combination on real datasets and report 1.3-5× QPS at 95% Recall@10 for arbitrary filters, which would matter for RAG and recommendation workloads if the numbers hold. That is the concrete contribution worth noting. The paper does a reasonable job of describing the three pieces as a single system rather than bolting them on separately. The exclusion-distance idea is straightforward to implement on top of standard HNSW and does not require per-filter indexes, so it keeps generality. The routing logic is also simple enough that it could be reproduced from the description. The main weakness is that the central mechanism still lacks supporting analysis. HNSW traversal depends on relative distance ordering for neighbor selection and entry-point movement. When distances are artificially increased for non-matching vectors, the greedy or beam search can follow different edges and miss candidates that would have been reached with unmodified distances. The abstract and the stress-test note both indicate no bounds on how far the distances can be pushed before recall on the filtered subset degrades, and no ablation that isolates the reshaping effect from the routing or the estimator. Without those checks, the reported QPS gains are hard to attribute cleanly to the new component. Experiments are described only at a high level, with no mention of exact baselines, variance across runs, or how recall was measured on the filtered subset alone. This paper is aimed at engineers who maintain vector indexes under mixed attribute filters and need something that works across selectivity ranges without heavy per-query tuning. A practitioner who already runs HNSW and wants a drop-in improvement for hybrid queries would find the routing and distance-reshaping ideas useful to try. The work is coherent on its own terms and engages the right prior literature, so it deserves a serious referee. I would send it to review and ask specifically for the missing traversal analysis and component ablations before accepting the performance claims.

Referee Report

2 major / 2 minor

Summary. The paper proposes FAVOR, a filter-agnostic ANNS method for hybrid vector-attribute queries. It integrates selectivity estimation with an HNSW-based inline filtering algorithm that uses an exclusion distance mechanism to dynamically reshape distances (pushing non-matching vectors away while promoting valid candidates) and a selectivity-driven search selector that routes low-selectivity queries to pre-filtering brute-force and others to optimized HNSW search. The central claim is that this yields 1.3-5× higher QPS at Recall@10 = 95% versus state-of-the-art methods for arbitrary filtering conditions while remaining competitive with tailored solutions.

Significance. If the performance claims and recall guarantees hold, FAVOR would offer a practical, general-purpose advance for filtered ANNS in production retrieval systems (recommendation, RAG) by avoiding the need for per-filter indexes or custom connectivity while delivering stable throughput across selectivity regimes. The unification of selectivity estimation and search routing is a clear engineering strength.

major comments (2)

[HNSW-based inline-filtering algorithm section] The exclusion-distance mechanism (described in the HNSW-based inline-filtering algorithm) lacks any analysis of its effect on HNSW traversal paths, greedy neighbor selection, or entry-point navigation. Because HNSW decisions depend on relative distance ordering, dynamically increasing distances for non-matching vectors can alter which edges are followed and potentially exclude valid filtered candidates that would have been reached under unmodified distances; no bound, ablation, or candidate-set-quality metric is provided to quantify this risk. This directly bears on the central QPS-at-95%-recall claim.
[Experiments section] The experimental evaluation reports 1.3-5× QPS gains at Recall@10 = 95% but supplies no details on the precise baselines (including their filtering implementations), error bars across runs, data-partitioning or exclusion rules, or the exact selectivity ranges tested. Without these, it is impossible to verify that the gains are attributable to the exclusion-distance and selector components rather than implementation artifacts or favorable dataset properties.

minor comments (2)

Notation for the exclusion distance and selectivity estimator should be introduced with explicit equations rather than prose descriptions to aid reproducibility.
Figure captions for the QPS-recall curves should explicitly state the filtering conditions (e.g., attribute cardinality, selectivity) used in each panel.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation and reproducibility.

read point-by-point responses

Referee: [HNSW-based inline-filtering algorithm section] The exclusion-distance mechanism (described in the HNSW-based inline-filtering algorithm) lacks any analysis of its effect on HNSW traversal paths, greedy neighbor selection, or entry-point navigation. Because HNSW decisions depend on relative distance ordering, dynamically increasing distances for non-matching vectors can alter which edges are followed and potentially exclude valid filtered candidates that would have been reached under unmodified distances; no bound, ablation, or candidate-set-quality metric is provided to quantify this risk. This directly bears on the central QPS-at-95%-recall claim.

Authors: We appreciate the referee highlighting this aspect. The exclusion distance is applied exclusively to non-matching vectors while leaving distances to matching vectors unchanged, thereby preserving relative orderings among all valid candidates and the original graph connectivity. This design guides greedy selection toward matching vectors without introducing new edges or altering the index. Nevertheless, we acknowledge that an explicit analysis of traversal dynamics would further substantiate the claims. In the revision we will add an ablation study reporting candidate-set quality metrics (e.g., fraction of valid vectors visited) and traversal-path statistics under reshaped versus original distances, directly quantifying any impact on recall. revision: yes
Referee: [Experiments section] The experimental evaluation reports 1.3-5× QPS gains at Recall@10 = 95% but supplies no details on the precise baselines (including their filtering implementations), error bars across runs, data-partitioning or exclusion rules, or the exact selectivity ranges tested. Without these, it is impossible to verify that the gains are attributable to the exclusion-distance and selector components rather than implementation artifacts or favorable dataset properties.

Authors: We agree that additional experimental details are required for full reproducibility and attribution of results. In the revised manuscript we will expand the Experiments section to specify: (i) exact baseline implementations and their filtering strategies (pre-filter, post-filter, or inline), (ii) error bars or standard deviations from repeated runs, (iii) data-partitioning and exclusion rules, and (iv) the precise selectivity ranges evaluated on each dataset. These clarifications will confirm that the reported QPS improvements arise from the proposed exclusion-distance and selector mechanisms. revision: yes

Circularity Check

0 steps flagged

No circularity: algorithmic proposal supported by experiments, not self-referential derivations

full rationale

The paper describes FAVOR as an integrated architecture combining selectivity estimation, an HNSW-based inline-filtering algorithm with an exclusion distance mechanism, and a selectivity-driven search selector. These are presented as novel algorithmic features without any equations, parameter fittings, or derivations that reduce to their own inputs by construction. Performance claims (1.3-5× QPS at Recall@10=95%) are grounded in extensive experiments on real-world datasets rather than mathematical self-consistency or self-citations. No load-bearing steps invoke uniqueness theorems, ansatzes smuggled via prior work, or renaming of known results. The method is self-contained as an engineering construction whose validity rests on empirical evaluation, not tautological reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, no explicit free parameters, axioms, or invented entities are stated. The exclusion distance is introduced as a new algorithmic mechanism rather than a fitted constant or postulated physical entity.

pith-pipeline@v0.9.0 · 5590 in / 1130 out tokens · 35162 ms · 2026-05-11T02:34:44.823348+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

[1]

Le and Tomás Mikolov

Quoc V. Le and Tomás Mikolov. 2014. Distributed Representations of Sentences and Documents. InICML, Vol. 32. JMLR.org, 1188–1196. http://proceedings.mlr.press/v32/le14.html

work page 2014
[2]

Martin Grohe. 2020. word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings of Structured Data. InPODS. ACM, 1–16. doi:10.1145/3375395.3387641

work page doi:10.1145/3375395.3387641 2020
[3]

Patrick S. H. Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. InNeurIPS. https://proceedings.neurips.cc/paper/2020/hash/ 6b493230205f780e1bc26945df7481...

work page 2020
[4]

Rae, Erich Elsen, and Laurent , Vol

Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, Diego de Las Casas, Aurelia Guy, Jacob Menick, Roman Ring, Tom Hennigan, Saffron Huang, Loren Maggiore, Chris Jones, Albin Cassirer, Andy Brock, Michela Paganini, Geoffrey Irving, Ori...

work page 2026
[5]

Jianguo Wang, Xiaomeng Yi, Rentong Guo, Hai Jin, Peng Xu, Shengjun Li, Xiangyu Wang, Xiangzhou Guo, Chengming Li, Xiaohai Xu, Kun Yu, Yuxing Yuan, Yinghao Zou, Jiquan Long, Yudong Cai, Zhenxiang Li, Zhifeng Zhang, Yihua Mo, Jun Gu, Ruiyi Jiang, Yi Wei, and Charles Xie. 2021. Milvus: A Purpose-Built Vector Data Management System. In SIGMOD. ACM, 2614–2627....

work page doi:10.1145/3448016.3457550 2021
[6]

Mengzhao Wang, Xiaoliang Xu, Qiang Yue, and Yuxiang Wang. 2021. A comprehensive survey and experimental comparison of graph-based approximate nearest neighbor search.PVLDB14, 11 (2021), 1964–1978. doi:10.14778/ 3476249.3476255

work page arXiv 2021
[7]

Yu A Malkov and Dmitry A Yashunin. 2018. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs.TPAMI42, 4 (2018), 824–836

work page 2018
[8]

Yuzheng Cai, Jiayang Shi, Yizhuo Chen, and Weiguo Zheng. 2024. Navigating Labels and Vectors: A Unified Approach to Filtered Approximate Nearest Neighbor Search.SIGMOD2, 6, Article 246 (2024), 27 pages. doi:10.1145/3698822

work page doi:10.1145/3698822 2024
[9]

Zhaoheng Li, Silu Huang, Wei Ding, Yongjoo Park, and Jianjun Chen. 2025. SIEVE: Effective Filtered Vector Search with Collection of Indexes.PVLDB18, 11 (2025), 4723–4736

work page 2025
[10]

Chaoji Zuo, Miao Qiao, Wenchao Zhou, Feifei Li, and Dong Deng. 2024. SeRF: Segment Graph for Range-Filtering Approximate Nearest Neighbor Search.SIGMOD2, 1 (2024), 69:1–69:26. doi:10.1145/3639324

work page doi:10.1145/3639324 2024
[11]

Liana Patel, Peter Kraft, Carlos Guestrin, and Matei Zaharia. 2024. ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data.SIGMOD2, 3 (2024), 120. doi:10.1145/3654923

work page doi:10.1145/3654923 2024
[12]

Guoyu Hu, Shaofeng Cai, Tien Tuan Anh Dinh, Zhongle Xie, Cong Yue, Gang Chen, and Beng Chin Ooi. 2025. HAKES: Scalable Vector Database for Embedding Search Service.PVLDB18, 9 (2025), 3049–3062. doi:10.14778/3746405.3746427

work page doi:10.14778/3746405.3746427 2025
[13]

Alexandr Andoni and Piotr Indyk. 2008. Near-optimal hashing algorithms for approximate nearest neighbor in high dimensions.Commun. ACM51, 1 (2008), 117–122. doi:10.1145/1327452.1327494

work page doi:10.1145/1327452.1327494 2008
[14]

Razenshteyn

Alexandr Andoni and Ilya P. Razenshteyn. 2015. Optimal Data-Dependent Hashing for Approximate Near Neighbors. InSTOC. 793–801. doi:10.1145/2746539.2746553

work page doi:10.1145/2746539.2746553 2015
[15]

Razenshteyn, and Ludwig Schmidt

Alexandr Andoni, Piotr Indyk, Thijs Laarhoven, Ilya P. Razenshteyn, and Ludwig Schmidt. 2015. Practical and Optimal LSH for Angular Distance. InNeurIPS. 1225–1233. https://proceedings.neurips.cc/paper/2015/hash/ 2823f4797102ce1a1aec05359cc16dd9-Abstract.html

work page 2015
[16]

Yury Elkin and Vitaliy Kurlin. 2022. Paired compressed cover trees guarantee a near linear parametrized complexity for all k-nearest neighbors search in an arbitrary metric space.CoRRabs/2201.06553 (2022). arXiv:2201.06553 https://arxiv.org/abs/2201.06553

work page arXiv 2022
[17]

Houle and Michael Nett

Michael E. Houle and Michael Nett. 2015. Rank-Based Similarity Search: Reducing the Dimensional Dependence. TPAMI37, 1 (2015), 136–150. doi:10.1109/TPAMI.2014.2343223

work page doi:10.1109/tpami.2014.2343223 2015
[18]

Kejing Lu, Hongya Wang, Wei Wang, and Mineichi Kudo. 2020. VHP: Approximate Nearest Neighbor Search via Virtual Hypersphere Partitioning.PVLDB13, 9 (2020), 1443–1455. doi:10.14778/3397230.3397240

work page doi:10.14778/3397230.3397240 2020
[19]

Marius Muja and David G. Lowe. 2014. Scalable Nearest Neighbor Algorithms for High Dimensional Data.TPAMI36, 11 (2014), 2227–2240. doi:10.1109/TPAMI.2014.2321376

work page doi:10.1109/tpami.2014.2321376 2014
[20]

Yury Malkov, Alexander Ponomarenko, Andrey Logvinov, and Vladimir Krylov. 2014. Approximate nearest neighbor algorithm based on navigable small world graphs.Inf. Syst.45 (2014), 61–68. doi:10.1016/J.IS.2013.10.006

work page doi:10.1016/j.is.2013.10.006 2014
[21]

Wei Wu, Junlin He, Yu Qiao, Guoheng Fu, Li Liu, and Jin Yu. 2022. HQANN: Efficient and Robust Similarity Search for Hybrid Queries with Structured and Unstructured Constraints. InCIKM. ACM, 4580–4584. doi:10.1145/3511808.3557610

work page doi:10.1145/3511808.3557610 2022
[22]

Mengzhao Wang, Lingwei Lv, Xiaoliang Xu, Yuxiang Wang, Qiang Yue, and Jiongkang Ni. 2023. An Efficient and Robust Framework for Approximate Nearest Neighbor Search with Attribute Constraint. InNeurIPS. http://papers. nips.cc/paper_files/paper/2023/hash/32e41d6b0a51a63a9a90697da19d235d-Abstract-Conference.html

work page 2023
[23]

Gaurav Gupta, Jonah Yi, Benjamin Coleman, Chen Luo, Vihan Lakshman, and Anshumali Shrivastava. 2023. CAPS: A Practical Partition Index for Filtered Similarity Search.CoRRabs/2308.15014 (2023). doi:10.48550/ARXIV.2308.15014 arXiv:2308.15014

work page doi:10.48550/arxiv.2308.15014 2023
[24]

Siddharth Gollapudi, Neel Karia, Varun Sivashankar, Ravishankar Krishnaswamy, Nikit Begwani, Swapnil Raz, Yiyong Lin, Yin Zhang, Neelam Mahapatro, Premkumar Srinivasan, Amit Singh, and Harsha Vardhan Simhadri. 2023. Filtered- DiskANN: Graph Algorithms for Approximate Nearest Neighbor Search with Filters. InWWW. ACM, 3406–3416. doi:10.1145/3543507.3583552

work page doi:10.1145/3543507.3583552 2023
[25]

Joshua Engels, Benjamin Landrum, Shangdi Yu, Laxman Dhulipala, and Julian Shun. 2024. Approximate Nearest Neighbor Search with Window Filters. InICML. OpenReview.net. https://openreview.net/forum?id=8t8zBaGFar

work page 2024
[26]

Yuexuan Xu, Jianyang Gao, Yutong Gou, Cheng Long, and Christian S. Jensen. 2024. iRangeGraph: Improvising Range- dedicated Graphs for Range-filtering Nearest Neighbor Search.SIGMOD2, 6 (2024), 239:1–239:26. doi:10.1145/3698814

work page doi:10.1145/3698814 2024
[27]

Anqi Liang, Pengcheng Zhang, Bin Yao, Zhongpu Chen, Yitong Song, and Guangxu Cheng. 2024. UNIFY: Unified Index for Range Filtered Approximate Nearest Neighbors Search.PVLDB18, 4 (2024), 1118–1130. https://www.vldb. , Vol. 1, No. 1, Article . Publication date: May 2026. FAVOR: Efficient Filter-Agnostic Vector ANNS Based on Selectivity-Aware Exclusion Dista...

work page 2024
[28]

Chaoji Zuo and Dong Deng. 2023. ARKGraph: All-Range Approximate K-Nearest-Neighbor Graph.PVLDB16, 10 (2023), 2645–2658. doi:10.14778/3603581.3603601

work page doi:10.14778/3603581.3603601 2023
[29]

Weijie Zhao, Shulong Tan, and Ping Li. 2022. Constrained Approximate Similarity Search on Proximity Graph.CoRR abs/2210.14958 (2022). doi:10.48550/ARXIV.2210.14958 arXiv:2210.14958

work page doi:10.48550/arxiv.2210.14958 2022
[30]

Gaurav Sehgal and Semih Salihoglu. 2025. NaviX: A Native Vector Index Design for Graph DBMSs With Robust Predicate-Agnostic Search Performance.PVLDB18, 11 (2025), 4438–4450. https://www.vldb.org/pvldb/vol18/p4438- sehgal.pdf

work page 2025
[31]

Jaromczyk and Godfried T

Jerzy W. Jaromczyk and Godfried T. Toussaint. 1992. Relative neighborhood graphs and their relatives.Proc. IEEE80, 9 (1992), 1502–1517. doi:10.1109/5.163414

work page doi:10.1109/5.163414 1992
[32]

1991.An introduction to probability theory and its applications, Volume 2

William Feller. 1991.An introduction to probability theory and its applications, Volume 2. Vol. 2. John Wiley & Sons

work page 1991
[33]

1993.Poisson Processes

JFC Kingman. 1993.Poisson Processes. Oxford University Press, United Kingdom

work page 1993
[34]

Jie Li, Haifeng Liu, Chuanghua Gui, Jianyu Chen, Zhenyun Ni, and Ning Wang. 2019. The Design and Implementation of a Real Time Visual Search System on JD E-commerce Platform. arXiv:1908.07389 [cs.IR]

work page arXiv 2019
[35]

Draper and Harry Smith

Norman R. Draper and Harry Smith. 1998.Applied Regression Analysis

work page 1998
[36]

Zeyu Wang, Haoran Xiong, Qitong Wang, Zhenying He, Peng Wang, Themis Palpanas, and Wei Wang. 2024. Dimensionality-Reduction Techniques for Approximate Nearest Neighbor Search: A Survey and Evaluation.IEEE Data Eng. Bull.48, 3 (2024), 63–80. http://sites.computer.org/debull/A24sept/p63.pdf

work page 2024
[37]

Tamer Özsu

Sairaj Voruganti and M. Tamer Özsu. 2025. MIRAGE-ANNS: Mixed Approach Graph-based Indexing for Approximate Nearest Neighbor Search.SIGMOD3, 3, Article 188 (2025), 27 pages. doi:10.1145/3725325

work page doi:10.1145/3725325 2025
[38]

Yun Peng, Byron Choi, Tsz Nam Chan, Jianye Yang, and Jianliang Xu. 2023. Efficient Approximate Nearest Neighbor Search in Multi-dimensional Databases.SIGMOD1, 1, Article 54 (2023), 27 pages. doi:10.1145/3588908

work page doi:10.1145/3588908 2023
[39]

Yun Peng, Byron Choi, Tsz Nam Chan, and Jianliang Xu. 2022. LAN: Learning-based Approximate k-Nearest Neighbor Search in Graph Databases. InICDE. 2508–2521. doi:10.1109/ICDE53745.2022.00233

work page doi:10.1109/icde53745.2022.00233 2022
[40]

Mengxu Jiang, Zhi Yang, Fangyuan Zhang, Guanhao Hou, Jieming Shi, Wenchao Zhou, Feifei Li, and Sibo Wang. 2025. DIGRA: A Dynamic Graph Indexing for Approximate Nearest Neighbor Search with Range Filter.SIGMOD3, 3, Article 148 (2025), 26 pages. doi:10.1145/3725399

work page doi:10.1145/3725399 2025
[41]

Mohsen Dehghankar and Abolfazl Asudeh. 2025. HENN: A Hierarchical Epsilon Net Navigation Graph for Approximate Nearest Neighbor Search. arXiv:2505.17368 [cs.DS] https://arxiv.org/abs/2505.17368

work page arXiv 2025
[42]

Jianyang Gao, Yutong Gou, Yuexuan Xu, Yongyi Yang, Cheng Long, and Raymond Chi-Wing Wong. 2025. Practical and Asymptotically Optimal Quantization of High-Dimensional Vectors in Euclidean Space for Approximate Nearest Neighbor Search.SIGMOD3, 3, Article 202 (2025), 26 pages. doi:10.1145/3725413

work page doi:10.1145/3725413 2025
[43]

Yutong Gou, Jianyang Gao, Yuexuan Xu, and Cheng Long. 2025. SymphonyQG: Towards Symphonious Integration of Quantization and Graph for Approximate Nearest Neighbor Search.SIGMOD3, 1, Article 80 (2025), 26 pages. doi:10.1145/3709730

work page doi:10.1145/3709730 2025
[44]

Chuangxian Wei, Bin Wu, Sheng Wang, Renjie Lou, Chaoqun Zhan, Feifei Li, and Yuanzhe Cai. 2020. AnalyticDB-V: A Hybrid Analytical Engine Towards Query Fusion for Structured and Unstructured Data.PVLDB13, 12 (2020), 3152–3165. doi:10.14778/3415478.3415541

work page doi:10.14778/3415478.3415541 2020
[45]

Yannis Chronis, Helena Caminal, Yannis Papakonstantinou, Fatma Özcan, and Anastasia Ailamaki. 2025. Filtered Vector Search: State-of-the-Art and Research Opportunities.PVLDB18, 12 (2025), 5488–5492. doi:10.14778/3750601.3750700

work page doi:10.14778/3750601.3750700 2025
[46]

Fangyuan Zhang, Mengxu Jiang, Guanhao Hou, Jieming Shi, Hua Fan, Wenchao Zhou, Feifei Li, and Sibo Wang. 2025. Efficient Dynamic Indexing for Range Filtered Approximate Nearest Neighbor Search.SIGMOD3, 3, Article 152 (2025), 26 pages. doi:10.1145/3725401

work page doi:10.1145/3725401 2025
[47]

Zhencan Peng, Miao Qiao, Wenchao Zhou, Feifei Li, and Dong Deng. 2025. Dynamic Range-Filtering Approximate Nearest Neighbor Search.PVLDB18, 10 (2025), 3256–3268. doi:10.14778/3748191.3748193 , Vol. 1, No. 1, Article . Publication date: May 2026

work page doi:10.14778/3748191.3748193 2025