Pith. sign in

REVIEW 3 major objections 6 minor 2 cited by

This paper claims that a unified tuning and filtering protocol can make filtered nearest-neighbor comparisons fair—and that on that protocol, the best method depends on the filter type: UNG for containment/equality, ACORN/DiskANN for overla

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

A systematic benchmark of filtered nearest-neighbor search algorithms shows no single winner: filter-then-search methods excel at containment and equality filters, while hybrid methods dominate overlap filters.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection The most substantial FANNS benchmark I've seen, worth citing and fixing, but the tuning protocol reuses test queries and lacks variance, so don't read the rankings as intrinsic yet. the 3 major comments →

arxiv 2509.07789 v1 pith:TONST737 submitted 2025-09-09 cs.DB

Filtered Approximate Nearest Neighbor Search: A Unified Benchmark and Systematic Experimental Study [Experiment, Analysis & Benchmark]

Jiayang Shi , Yuzheng Cai , Weiguo Zheng This is my paper
classification cs.DB
keywords filtered approximate nearest neighbor searchbenchmarkparameter tuningfilter-then-searchsearch-then-filterhybrid-searchrecall-QPS trade-offgraph-based index
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper's thesis is that existing comparisons of Filtered Approximate Nearest Neighbor Search (FANNS) methods are unreliable, because each method is tuned differently, tested on different datasets and workloads, and reported with different experimental choices. It tries to establish that a common evaluation protocol—grouping methods into filter-then-search, search-then-filter, and hybrid-search, tuning all methods with one parameter-selection procedure, and using one bitset-based filtering implementation—can isolate each paradigm's genuine speed-accuracy trade-offs. Under that protocol, the paper reports that no single method dominates: UNG wins containment and equality queries, ACORN and Stitched DiskANN lead on overlap queries, and post-filtering methods stay competitive at low k but degrade at high k. The benchmark also finds that query-label length and filter selectivity move performance more than dataset scale does. If the protocol is sound, it gives practitioners a defensible basis for choosing an algorithm per workload instead of relying on isolated published numbers.

Core claim

The paper's central claim is that a fair, parameter-aware evaluation framework can separate intrinsic algorithm trade-offs from tuning artifacts in filtered nearest neighbor search. It groups methods into filter-then-search (UNG, ACORN, brute-force), search-then-filter (post-filter HNSW, IVFPQ), and hybrid-search (Filtered-DiskANN, Stitched-DiskANN, NHQ, CAPS), and applies one tuning protocol plus one bitset-filtering implementation across six labeled datasets and four filter semantics. It reports that the best method depends on the filter constraint: UNG dominates containment and equality because its label graph never expands into invalid nodes; ACORN-1 and Stitched-DiskANN lead on overlap,

What carries the argument

The load-bearing mechanism is the paper's unified tuning protocol (Algorithm 1): it partitions the parameter space into subspaces, builds small indexes on a random sample, scores each configuration by interpolated QPS at fixed recall targets, and selects one representative configuration per subspace for full-size evaluation. Combined with a standardized bitset-filter implementation (bitwise AND/OR over precomputed label-to-vector bitsets) and the three-way taxonomy, this pipeline turns ad-hoc hyperparameter choices into a reproducible procedure, so differences in QPS-recall frontiers can be attributed to algorithm design rather than tuning luck.

Load-bearing premise

The framework's fairness rests on the assumption that parameters selected on a small random sample of each dataset, then frozen, are still near-optimal for the full index; the evaluation does not hold out a separate query set to test that assumption.

What would settle it

Re-run the benchmark with a strict train/test split: tune parameters on one random half of each dataset's queries, then evaluate on the held-out half. If the reported rankings or Pareto frontiers change materially—for example, if a method the paper ranks low reaches the same recall at higher QPS under held-out tuning—the framework is not isolating algorithm merit. A simpler check: for one dataset, compare the sample-tuned parameters against a full grid search on the entire index; identical QPS-recall frontiers would support the protocol, divergence would refute it.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • For containment and equality workloads, UNG is the recommended default; its throughput stays high even as query labels lengthen, because the label graph restricts search to a small valid subgraph.
  • For overlap workloads, ACORN-1 and Stitched DiskANN offer the best balance; UNG degrades as label complexity grows.
  • Post-filter IVFPQ is a poor choice at high k, because valid items scatter across inverted lists; prefer UNG/ACORN for containment/equality and DiskANN/ACORN for overlap at recall@100.
  • Dataset size is not the main pressure: going from 10% to 100% of a dataset changes QPS-recall curves only slightly; label selectivity and query-label length matter more.
  • Methods that need base labels during construction (UNG, DiskANN variants, NHQ, CAPS) index more slowly; ACORN-1, HNSW, and IVFPQ build quickly and tolerate label changes, an operational constraint for dynamic data.

Where Pith is reading between the lines

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

  • We infer the tuning protocol could become a community standard: future FANNS papers reporting only hand-tuned numbers would be hard to compare against this benchmark, though the authors stop short of saying so.
  • We infer the rankings generalize only as far as the six real-world label distributions represent production workloads; heavily skewed, correlated, or hierarchical labels could shift the rankings.
  • Because the paper tests only in-memory indexes, we infer rankings may change on disk-resident billion-scale data, where I/O access patterns and graph reordering become first-order factors.
  • The framework appears compatible with numerical range filters, which the paper explicitly leaves out; we infer the same unified tuning pipeline could apply to those methods, but that is untested.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper presents a benchmark and systematic experimental study of Filtered Approximate Nearest Neighbor Search (FANNS). It proposes a taxonomy (filter-then-search, search-then-filter, hybrid-search), a parameter tuning framework (Algorithm 1) that samples a subset of the dataset, sweeps parameter configurations, interpolates QPS at fixed recall targets, and selects representative parameters per subspace. It evaluates 9 algorithms on 6 real-world datasets across four filter scenarios (containment, equality, overlap, fixed-length equality), tests effects of query label length, selectivity, top-k, and dataset size, and provides recommendations. The central claim is that the framework isolates and accurately reflects fundamental performance characteristics and core trade-offs of each algorithmic paradigm. The paper also releases code and datasets.

Significance. If the central claim holds, this would be a valuable reference for FANNS algorithm selection and a standardized benchmark. The scope is substantial: 41,000 parameter combinations, ~1,300 full indexes, 6 datasets, multiple scenarios. The paper includes machine-checkable artifacts (open-source code), a clear taxonomy, and a structured tuning protocol, which are strengths. However, the significance depends on the validity of the tuning/evaluation protocol, which has potential issues (tuning on a sample and using the same query set for evaluation, no variance reporting).

major comments (3)
  1. [Algorithm 1 (lines 5-16) and Section 5.3] The tuning protocol uses a randomly sampled subset D' to select parameters, but the selected parameters are then used to build full indexes and evaluate on the same query set Q with no validation on a held-out portion of the data. Since Algorithm 1 explicitly averages over the full query set Q (line 10) and selects the best-performing configuration per subspace (line 16), the reported QPS-recall frontiers may reflect overfitting to the tuning queries. The paper provides no transfer check (e.g., comparing tuned parameters on D' vs. full D, or evaluating on a disjoint query set) and no variance/error bars. This is load-bearing because the central claim in Section 1.2 is that the framework 'isolates and accurately reflect[s] the fundamental performance characteristics and core trade-offs of each algorithmic paradigm.' I request either (a) a held-out query set for final evaluation, (b) a dem
  2. [Algorithm 1 and Section 5.2] The transfer of parameters tuned on a small random sample D' to full-size indexes is assumed but not validated. Graph-based methods (HNSW M/ef, DiskANN R/L, ACORN γ) and UNG have size-dependent optimal parameters; the paper does not report the sample size |D'| for each dataset or any experiment showing that the relative ranking of configurations is preserved between D' and the full dataset. Without such a check, the comparisons may reflect tuning artifacts rather than intrinsic algorithm trade-offs. Please provide the sample sizes and a correlation/stability analysis, or justify the assumption with evidence from the 1,300 full-index builds.
  3. [Section 5 and Figure 9] No error bars, multiple runs, or statistical significance tests are reported for any QPS or recall measurement. Since the paper makes comparative claims (e.g., 'UNG is well-suited for containment and equality', 'DiskANN performs well in overlap'), the stability of the rankings is unknown. I am not asking for full repetition of all 41,000 configurations, but at least for a subset (e.g., main Pareto-frontier configurations on representative datasets) to assess run-to-run variance and establish that the observed margins are not noise.
minor comments (6)
  1. [Section 1.1, Table 1] The table lists parameter-space dimensions, but the specific parameter names and ranges for each algorithm are not given in the text. A supplementary table or appendix listing the subspaces and parameter grids would improve reproducibility.
  2. [Section 4, Algorithm 1] The recall targets R are defined as a list (e.g., 0.8, 0.9, 0.95), but the interpolation method (linear in what space?) is not fully specified. Clarify whether interpolation is on raw recall/QPS or log-transformed values, and how ties are broken in ranking.
  3. [Section 5.1, Table 3] The dataset sizes for TripClick and LAION1M are listed as ~1M, but for YouTube-Audio the description says '5,000,000' while later Section 5.1 text mentions 'our largest dataset is limited to 500M vectors' — this appears to be a typo (likely 5M vs 500M). Please correct.
  4. [Section 5.4] The fixed-length equality synthetic data generation says 'each position has three possible values, chosen with equal probability' and selectivity is 1/81, but with 4 positions and 3 values there are 3^4=81 combinations, each with probability (1/3)^4=1/81. This is consistent, but the sentence 'we followed the implementation in NHQ paper' should cite a specific section or hyperparameter (e.g., number of labels per point).
  5. [Section 5.6] The selectivity groups are described as '75th, 50th, 25th, and 1st percentiles' — this is ambiguous. Do these refer to percentiles of the selectivity distribution? Clarify the direction (e.g., 'low selectivity' = 1st percentile) and how the query groups are constructed.
  6. [Section 6, Recommendations] The recommendations are useful but some are not directly tied to the figures (e.g., 'avoid post-filter IVFPQ at high k' is supported by Figure 13, but the statement about ACORN-1 and stitch-diskann in overlap could cite Figure 11 or 12).

Circularity Check

1 steps flagged

Tuning and evaluation share the same query set; reported QPS-recall frontiers are in-sample fits, so the 'accurate reflection' claim is partially circular.

specific steps
  1. fitted input called prediction [Algorithm 1 (lines 5–17) and Figure 6 caption; Section 1.2 central claim; Section 5.3 evaluation]
    "Generate a sampled dataset D′←RandomSample(D) ... {qps_s, recall_s}←PerformSearch(A,I,s,Q) ... θ∗←arg min θ∈Θ_i rθ ... Result←Result∪{θ∗} ... 'Consequently, our benchmarking results isolate and accurately reflect the fundamental performance characteristics and core trade-offs of each algorithmic paradigm.'"

    Algorithm 1 optimizes each configuration θ on the full query set Q (line 10), then selects the representative θ* with the best interpolated QPS at fixed recall targets on Q. All reported QPS-recall frontiers in Section 5.3 onward are measured on the same Q; no held-out query split is described. The reported frontier for each algorithm is therefore the selected optimum of the tuning objective on Q, not an independent estimate. The central claim that the benchmark 'isolate[s] and accurately reflect[s] fundamental performance characteristics' reduces to an in-sample fit: the rankings are generated by the same queries used to choose the parameters that produce those rankings.

full rationale

The paper is an experimental benchmark, not a mathematical derivation, and most of its content is independent empirical measurement of nine external algorithms. The main circular component is in the parameter-tuning protocol: the same query set Q is used both to select representative parameters (Algorithm 1 line 10) and to report the final QPS-recall curves (Section 5.3 onward). Because θ* is chosen by maximizing interpolated QPS at fixed recall levels on Q, the reported frontiers are in-sample evaluations of parameters fitted to those queries. The Section 1.2 assertion that the results 'isolate and accurately reflect' fundamental trade-offs is therefore not supported by any out-of-sample check; it is an extrapolation from the tuning set. This is a genuine fitted-input/called-prediction pattern, though it is partial: the benchmark still measures real algorithm behavior, and the tuning is applied uniformly. The authors' self-citation to their own UNG paper [3] is not itself circular, since UNG's performance is measured directly in this work; the elevated score is due to the query-set reuse, not to self-citation. The paper also reports no variance or error bars, so stability of the resulting rankings cannot be assessed; that is a reporting gap rather than a circular step.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The central claims rest primarily on experimental design choices (recall targets, subspaces, sample sizes, query sampling) and on the assumption that the authors' implementations of both their own and external algorithms are unbiased. No new mathematical entities are introduced.

free parameters (5)
  • Recall targets R for interpolation = [0.8, 0.9, 0.95]
    Algorithm 1 line 12 fixes these recall levels to rank parameter configurations; chosen by hand, not derived from data.
  • Subspace partition of parameter space
    Algorithm 1 line 1 divides the parameter space into subspaces, but the partition method and number of subspaces are unspecified, affecting representativeness.
  • Random sample size D'
    Algorithm 1 line 5 samples a subset for tuning; the sample size is not reported, making the tuning procedure non-reproducible.
  • Query count per difficulty group = 1000
    Section 5.5 randomly selects 1,000 queries per group; this sample size influences variance but no confidence intervals are given.
  • Fixed-length equality synthetic label parameters = length 4, 3 values per position
    Section 5.4 generates synthetic labels following the NHQ paper; these parameters define selectivity (1/81) and affect the fixed-length equality comparison.
axioms (4)
  • domain assumption QPS and Recall@k are the appropriate performance metrics for FANNS evaluation
    The paper uses these metrics throughout Section 2.2; they are standard for ANNS but shape all reported trade-offs.
  • domain assumption The selected datasets and generated query labels are representative of real-world FANNS workloads
    Section 5.1 describes six real-world datasets and query sampling; representativeness is assumed, not demonstrated.
  • domain assumption The authors' implementations and modifications of external algorithms faithfully represent the original algorithms
    Section 5.1 states multi-label support was added to Filtered-DiskANN and Stitched-DiskANN without altering graph construction; faithful representation is required for fair comparison.
  • ad hoc to paper Parameter configurations tuned on a random sample of the dataset transfer to full-size indexes
    Algorithm 1 selects parameters on sampled subset D' and then builds full indexes; the transferability is assumed and not validated on a held-out portion.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Filtered Approximate Nearest Neighbor Search: A Unified Benchmark and Systematic Experimental Study [Experiment, Analysis & Benchmark]." pith.science (2026). https://pith.science/paper/TONST737

@misc{pith2026250907789,
  author       = {Pith},
  title        = {Pith review of: Filtered Approximate Nearest Neighbor Search: A Unified Benchmark and Systematic Experimental Study [Experiment, Analysis & Benchmark]},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TONST737}},
  note         = {Machine review of arXiv:2509.07789}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

For a given dataset $\mathcal{D}$ and structured label $f$, the goal of Filtered Approximate Nearest Neighbor Search (FANNS) algorithms is to find top-$k$ points closest to a query that satisfy label constraints, while ensuring both recall and QPS (Queries Per Second). In recent years, many FANNS algorithms have been proposed. However, the lack of a systematic investigation makes it difficult to understand their relative strengths and weaknesses. Additionally, we found that: (1) FANNS algorithms have coupled, dataset-dependent parameters, leading to biased comparisons. (2) Key impact factors are rarely analyzed systematically, leaving unclear when each algorithm performs well. (3) Disparate datasets, workloads, and biased experiment designs make cross-algorithm comparisons unreliable. Thus, a comprehensive survey and benchmark for FANNS is crucial to achieve the following goals: designing a fair evaluation and clarifying the classification of algorithms, conducting in-depth analysis of their performance, and establishing a unified benchmark. First, we propose a taxonomy (dividing methods into \textit{filter-then-search}, \textit{search-then-filter}, \textit{hybrid-search}) and a systematic evaluation framework, integrating unified parameter tuning and standardized filtering across algorithms to reduce implementation-induced performance variations and reflect core trade-offs. Then, we conduct a comprehensive empirical study to analyze how query difficulty and dataset properties impact performance, evaluating robustness under pressures like filter selectivity, Recall@k, and scalability to clarify each method's strengths. Finally, we establish a standardized benchmark with real-world datasets and open-source related resources to ensure reproducible future research.

Figures

Figures reproduced from arXiv: 2509.07789 by Jiayang Shi, Weiguo Zheng, Yuzheng Cai.

Figure 2
Figure 2. Figure 2: Example of ACORN-1: it illustrates that ACORN [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of UNG. The dataset is partitioned into [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example of Stitched DiskANN. query, we perform an initial ANNS search on this index with a search parameter 𝑙 that controls the size of results. This result queue is then scanned to check for filter satisfaction. If at least 𝑘 valid results are found, the top-𝑘 among them are returned. If the count is insufficient, the search is re-issued with an expanded scope (e.g., by doubling 𝑙) and the process is repe… view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of Parameter Tuning Algorithm. The algorithm proceeds as follows: 1) randomly sample some points [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distributions of base label, size of base label sets, and the number of groups. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Scatter plots showing the effect of varying param [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Query performance on 6 real-world datasets [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Result of Fixed Length Equality Scenario. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Results of Varying Query Label Length [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Results of Varying Query Selectivity. points after filtering is small, thus limiting the number of distance computations and maintaining a relatively high QPS. Similarly, the performance of ACORN-𝛾 often resembles the brute-force method, which can be attributed to cases where low selectivity effectively prunes the search space to a small number of candidates. In this scenario, as query labels grow longer,… view at source ↗
Figure 13
Figure 13. Figure 13: Results of Varying Top-k [PITH_FULL_IMAGE:figures/full_fig_p011_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Results of Varying Base Dataset Size. the selection of top-k and its accuracy [12]. Thus, in this set of experiments, we evaluate the impact of varying the value of top-k on the performance of different algorithms. For each scenario, we test a range of top-k values, including 𝑘 = 1, 𝑘 = 25, 𝑘 = 50, and 𝑘 = 100. We guarantee that the selected queries always have at least 𝑘 ground truth results, regardless … view at source ↗
Figure 15
Figure 15. Figure 15: Effect of Varying Base Datasets Size on index con [PITH_FULL_IMAGE:figures/full_fig_p012_15.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Query-aware Routing for Filtered Approximate Nearest Neighbors Search

    cs.DB 2026-06 unverdicted novelty 7.0

    A machine-learned router predicts per-query recall for filtered ANN methods and selects the recall-QPS optimal one, outperforming fixed baselines on five unseen datasets.

  2. Simple and Fast Algorithm for Graph-based Filtered Approximate Nearest Neighbor Search (Full Version)

    cs.DB 2026-07 conditional novelty 5.0

    A labeled flat proximity graph built from full- and single-attribute partitions supports arbitrary filtered ANNS and beats UNG on 1–2 attribute queries at similar recall.

Reference graph

Works this paper leans on

54 extracted references · 23 canonical work pages · cited by 2 Pith papers · 1 internal anchor

  1. [1]

    Sami Abu-El-Haija, Nisarg Kothari, Joonseok Lee, Paul Natsev, George Toderici, Balakrishnan Varadarajan, and Sudheendra Vijayanarasimhan. 2016. YouTube- 8M: A Large-Scale Video Classification Benchmark. arXiv:1609.08675 [cs.CV] https://arxiv.org/abs/1609.08675

    Pith/arXiv arXiv 2016

  2. [2]

    Marián Boguñá, Dmitri Krioukov, and K. C. Claffy. 2008. Navigability of complex networks.Nature Physics5, 1 (Nov. 2008), 74–80. https://doi.org/10.1038/ nphys1130

    work page 2008

  3. [3]

    Yuzheng Cai, Jiayang Shi, Yizhuo Chen, and Weiguo Zheng. 2024. Navigat- ing Labels and Vectors: A Unified Approach to Filtered Approximate Nearest Neighbor Search.Proc. ACM Manag. Data2, 6, Article 246 (Dec. 2024), 27 pages. https://doi.org/10.1145/3698822

    doi:10.1145/3698822 2024

  4. [4]

    Benjamin Coleman, Santiago Segarra, Anshumali Shrivastava, and Alex Smola. 2021. Graph Reordering for Cache-Efficient Near Neighbor Search. arXiv:2104.03221 [cs.DS] https://arxiv.org/abs/2104.03221

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  5. [5]

    Wei Dong, Charikar Moses, and Kai Li. 2011. Efficient k-nearest neighbor graph construction for generic similarity measures. InProceedings of the 20th International Conference on World Wide Web(Hyderabad, India)(WWW ’11). Association for Computing Machinery, New York, NY, USA, 577–586. https: //doi.org/10.1145/1963405.1963487

    arXiv 2011

  6. [6]

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2025. The Faiss library. arXiv:2401.08281 [cs.LG] https://arxiv.org/abs/2401.08281

    Pith/arXiv arXiv 2025

  7. [7]

    Karima Echihabi, Kostas Zoumpatianos, and Themis Palpanas. 2021. New trends in high-d vector similarity search: al-driven, progressive, and distributed.Pro- ceedings of the VLDB Endowment14, 12 (2021), 3198–3201

    work page 2021

  8. [8]

    [n.d.].Elasticsearch

    Elastic. [n.d.].Elasticsearch. https://github.com/elastic/elasticsearch

    work page

  9. [9]

    Cong Fu, Chao Xiang, Changxu Wang, and Deng Cai. 2025. Fast Approx- imate Nearest Neighbor Search With The Navigating Spreading-out Graph. arXiv:1707.00143 [cs.LG] https://arxiv.org/abs/1707.00143

    Pith/arXiv arXiv 2025

  10. [10]

    Jianyang Gao, Yutong Gou, Yuexuan Xu, Yongyi Yang, Cheng Long, and Raymond Chi-Wing Wong. 2024. Practical and Asymptotically Optimal Quantization of High-Dimensional Vectors in Euclidean Space for Approximate Nearest Neighbor Search. arXiv:2409.09913 [cs.DB] https://arxiv.org/abs/2409.09913

    Pith/arXiv arXiv 2024

  11. [11]

    Jianyang Gao and Cheng Long. 2024. RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search.Proc. ACM Manag. Data2, 3, Article 167 (May 2024), 27 pages. https: //doi.org/10.1145/3654970

    doi:10.1145/3654970 2024

  12. [12]

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, and Haofen Wang. 2023. Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997(2023)

    Pith/arXiv arXiv 2023

  13. [13]

    Tiezheng Ge, Kaiming He, Qifa Ke, and Jian Sun. 2014. Optimized Product Quantization.IEEE Transactions on Pattern Analysis and Machine Intelligence36, 4 (2014), 744–755. https://doi.org/10.1109/TPAMI.2013.240

    doi:10.1109/tpami.2013.240 2014

  14. [14]

    Siddharth Gollapudi, Neel Karia, Varun Sivashankar, Ravishankar Krishnaswamy, Nikit Begwani, Swapnil Raz, Yiyong Lin, Yin Zhang, Neelam Mahapatro, Premku- mar Srinivasan, et al. 2023. Filtered-diskann: Graph algorithms for approximate nearest neighbor search with filters. InProceedings of the ACM Web Conference

    2023

  15. [15]

    2018.constrainedANN

    Gaurav Gupta. 2018.constrainedANN. https://github.com/gaurav16gupta/ constrainedANN

    work page 2018

  16. [16]

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

    Pith/arXiv arXiv 2023

  17. [17]

    Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. 2013. Learning deep structured semantic models for web search using clickthrough data. InProceedings of the 22nd ACM International Conference on Information & Knowledge Management(San Francisco, California, USA)(CIKM ’13). Association for Computing Machinery, New York, NY, USA, ...

    work page arXiv 2013

  18. [18]

    Wenqi Jiang, Shigang Li, Yu Zhu, Johannes De Fine Licht, Zhenhao He, Runbin Shi, Cedric Renggli, Shuai Zhang, Theodoros Rekatsinas, Torsten Hoefler, and Gustavo Alonso. 2023. Co-design Hardware and Algorithm for Vector Search. InProceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis(Denver, CO, USA)(SC...

    arXiv 2023

  19. [20]

    Herve Jégou, Matthijs Douze, and Cordelia Schmid. 2011. Product Quantization for Nearest Neighbor Search.IEEE Transactions on Pattern Analysis and Machine Intelligence33, 1 (2011), 117–128. https://doi.org/10.1109/TPAMI.2010.57

    doi:10.1109/tpami.2010.57 2011

  20. [21]

    2022.TKDE-under-review-Native-Hybrid-Queries-via-ANNS

    KGLab-HDU. 2022.TKDE-under-review-Native-Hybrid-Queries-via-ANNS. https://github.com/KGLab-HDU/TKDE-under-review-Native-Hybrid- Queries-via-ANNS

    work page 2022

  21. [22]

    Jon Kleinberg. 2000. Kleinberg, J. Navigation in a small world. Nature 406, 845. Nature406 (09 2000), 845. https://doi.org/10.1038/35022643

    work page doi:10.1038/35022643 2000

  22. [23]

    Vihan Lakshman, ChoonHui Teo, Xiaowen Chu, Priyanka Nigam, Abhinandan Patni, Pooja Maknikar, and SVN Vishwanathan. [n.d.]. Embracing Structure in Data for Billion-Scale Semantic Product Search. ([n. d.])

    work page

  23. [24]

    Chao Li, Zhiyuan Liu, Mengmeng Wu, Yuchi Xu, Huan Zhao, Pipei Huang, Guoliang Kang, Qiwei Chen, Wei Li, and Dik Lun Lee. 2019. Multi-Interest Network with Dynamic Routing for Recommendation at Tmall. InProceedings of the 28th ACM International Conference on Information and Knowledge Management (Beijing, China)(CIKM ’19). Association for Computing Machiner...

    work page arXiv 2019

  24. [25]

    Sen Li, Fuyu Lv, Taiwei Jin, Guli Lin, Keping Yang, Xiaoyi Zeng, Xiao-Ming Wu, and Qianli Ma. 2021. Embedding-based Product Retrieval in Taobao Search. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. https://doi.org/10.1145/3447548.3467101

    arXiv 2021

  25. [26]

    Wen Li, Ying Zhang, Yifang Sun, Wei Wang, Mingjie Li, Wenjie Zhang, and Xuemin Lin. 2019. Approximate nearest neighbor search on high dimensional data—experiments, analyses, and improvement.IEEE Transactions on Knowledge and Data Engineering32, 8 (2019), 1475–1488

    2019

  26. [27]

    Anqi Liang, Pengcheng Zhang, Bin Yao, Zhongpu Chen, Yitong Song, and Guangxu Cheng. 2025. UNIFY: Unified Index for Range Filtered Approximate Nearest Neighbors Search. arXiv:2412.02448 [cs.DS] https://arxiv.org/abs/2412. 02448

    Pith/arXiv arXiv 2025

  27. [28]

    Alessandro Magnani, Feng Liu, Suthee Chaidaroon, Sachin Yadav, Praveen Reddy Suram, Ajit Puthenputhussery, Sijie Chen, Min Xie, Anirudh Kashi, Tony Lee, and Ciya Liao. 2022. Semantic Retrieval at Walmart. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Washington DC, USA)(KDD ’22). Association for Computing Machinery...

    arXiv 2022

  28. [29]

    Yu A Malkov and Dmitry A Yashunin. 2018. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs.IEEE transactions on pattern analysis and machine intelligence42, 4 (2018), 824–836

    2018

  29. [30]

    Malteos. 2022. Aspect Paper Embeddings. https://huggingface.co/datasets/ malteos/aspect-paper-embeddings

    work page 2022

  30. [31]

    Jiongkang Ni, Xiaoliang Xu, Yuxiang Wang, Can Li, Jiajie Yao, Shihai Xiao, and Xuecang Zhang. 2023. DiskANN++: Efficient Page-based Search over Isomorphic Mapped Graph Index using Query-sensitivity Entry Vertex. arXiv:2310.00402 [cs.IR] https://arxiv.org/abs/2310.00402

    Pith/arXiv arXiv 2023

  31. [32]

    Priyanka Nigam, Yiwei Song, Vijai Mohan, Vihan Lakshman, Weitian (Allen) Ding, Ankit Shingavi, Choon Hui Teo, Hao Gu, and Bing Yin. 2019. Semantic Product Search. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. https://doi.org/10.1145/3292500.3330759

    arXiv 2019

  32. [33]

    Xichuan Niu, Bofang Li, Chenliang Li, Rong Xiao, Haochuan Sun, Honggang Wang, Hongbo Deng, and Zhenzhong Chen. 2020. Gated Heterogeneous Graph Representation Learning for Shop Search in E-commerce. InProceedings of the 29th ACM International Conference on Information & Knowledge Management (Virtual Event, Ireland)(CIKM ’20). Association for Computing Mach...

    work page arXiv 2020

  33. [34]

    OpenMP Architecture Review Board. 2008. OpenMP Application Program Inter- face Version 3.0. http://www.openmp.org/mp-documents/spec30.pdf

    2008

  34. [35]

    Malte Ostendorff, Till Blume, Terry Ruas, Bela Gipp, and Georg Rehm. 2022. Specialized document embeddings for aspect-based similarity of research papers. InProceedings of the 22nd ACM/IEEE Joint Conference on Digital Libraries. 1–12

    work page 2022

  35. [36]

    Liana Patel. 2024. ACORN. https://github.com/stanford-futuredata/ACORN

    work page 2024

  36. [37]

    Liana Patel, Peter Kraft, Carlos Guestrin, and Matei Zaharia. 2024. ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Struc- tured Data.Proc. ACM Manag. Data2, 3, Article 120 (May 2024), 27 pages. https://doi.org/10.1145/3654923

    doi:10.1145/3654923 2024

  37. [38]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision.CoRRabs/2103.00020 (2021). arXiv:2103.00020 https://arxiv.org/abs/2103.00020

    Pith/arXiv arXiv 2021

  38. [39]

    Navid Rekabsaz, Oleg Lesota, Markus Schedl, Jon Brassey, and Carsten Eickhoff

  39. [40]

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki

  40. [41]

    Harsha Vardhan Simhadri, Martin Aumüller, Amir Ingber, Matthijs Douze, George Williams, Magdalen Dobson Manohar, Dmitry Baranchuk, Edo Liberty, Frank Liu, Ben Landrum, Mazin Karjikar, Laxman Dhulipala, Meng Chen, Yue Chen, Rui Ma, Kai Zhang, Yuzheng Cai, Jiayang Shi, Yizhuo Chen, Weiguo Zheng, Zihao Wan, Jie Yin, and Ben Huang. 2024. Results of the Big AN...

    Pith/arXiv arXiv 2024

  41. [42]

    arXiv preprint arXiv:2111.02114(2021)

    LAION-400M: Open Dataset of Clip-filtered 400 Million Image-text Pairs. arXiv preprint arXiv:2111.02114(2021)

    Pith/arXiv arXiv 2021

  42. [43]

    Jitendra Nath Singh and Sanjay K. Dwivedi. 2015. Performance Evaluation of Search Engines Using Enhanced Vector Space Model.Journal of Computer Science 11, 4 (Jul 2015), 692–698. https://doi.org/10.3844/jcssp.2015.692.698

    work page doi:10.3844/jcssp.2015.692.698 2015

  43. [44]

    Harsha Vardhan Simhadri, Ravishankar Krishnaswamy, Gopal Srinivasa, Suhas Jayaram Subramanya, Andrija Antonijevic, Dax Pryce, David Kaczyn- ski, Shane Williams, Siddarth Gollapudi, Varun Sivashankar, Neel Karia, Aditi Singh, Shikhar Jaiswal, Neelam Mahapatro, Philip Adams, Bryan Tower, and Yash Patel. [n.d.]

    work page

  44. [45]

    2019.DiskANN: fast accurate billion- point nearest neighbor search on a single node

    Suhas Jayaram Subramanya, Devvrit, Rohan Kadekodi, Ravishankar Kr- ishaswamy, and Harsha Vardhan Simhadri. 2019.DiskANN: fast accurate billion- point nearest neighbor search on a single node. Curran Associates Inc., Red Hook, NY, USA

    work page 2019

  45. [46]

    SuhasJayaram Subramanya, Fnu Devvrit, HarshaVardhan Simhadri, Ravishankar Krishnawamy, and Rohan Kadekodi. 2019. DiskANN: Fast Accurate Billion- point Nearest Neighbor Search on a Single Node.Neural Information Processing Systems,Neural Information Processing Systems(Nov 2019)

    work page 2019

  46. [47]

    Mengzhao Wang, Lingwei Lv, Xiaoliang Xu, Yuxiang Wang, Qiang Yue, and Jiongkang Ni. 2024. An efficient and robust framework for approximate near- est neighbor search with attribute constraint.Advances in Neural Information Processing Systems36 (2024)

    work page 2024

  47. [48]

    Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li

    Bart Thomee, David A. Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li. 2016. YFCC100M: the new data in multimedia research.Commun. ACM59, 2 (Jan. 2016), 64–73. https: //doi.org/10.1145/2812802

    doi:10.1145/2812802 2016

  48. [49]

    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. InProceedings of the 31st ACM International Conference on Information & Knowledge Management. 4580–4584

    work page 2022

  49. [50]

    Mengzhao Wang, Xiaoliang Xu, Qiang Yue, and Yuxiang Wang. 2021. A com- prehensive survey and experimental comparison of graph-based approximate nearest neighbor search.arXiv preprint arXiv:2101.12631(2021)

    Pith/arXiv arXiv 2021

  50. [51]

    Penghao Zhao, Hailin Zhang, Qinhan Yu, Zhengren Wang, Yunteng Geng, Fangcheng Fu, Ling Yang, Wentao Zhang, and Bin Cui. 2024. Retrieval- Augmented Generation for AI-Generated Content: A Survey.arXiv preprint arXiv:2402.19473(2024)

    Pith/arXiv arXiv 2024

  51. [52]

    Tiannuo Yang, Wen Hu, Wangqi Peng, Yusen Li, Jianguo Li, Gang Wang, and Xiaoguang Liu. 2024. VDTuner: Automated Performance Tuning for Vector Data Management Systems. In2024 IEEE 40th International Conference on Data Engineering (ICDE). 4357–4369. https://doi.org/10.1109/ICDE60146.2024.00332

    arXiv 2024

  52. [53]

    Chaoji Zuo, Miao Qiao, Wenchao Zhou, Feifei Li, and Dong Deng. 2024. SeRF: Segment Graph for Range-Filtering Approximate Nearest Neighbor Search.Proc. ACM Manag. Data2, 1, Article 69 (March 2024), 26 pages. https://doi.org/10. 1145/3639324 14

    work page 2024

  53. [54]

    Chaoji Zuo and Dong Deng. 2023. ARKGraph: All-Range Approximate K-Nearest- Neighbor Graph.Proc. VLDB Endow.16, 10 (June 2023), 2645–2658. https: //doi.org/10.14778/3603581.3603601

    work page arXiv 2023

  54. [2021]

    InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval

    Tripclick: The Log Files of a Large Health Web Search Engine. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2507–2513

    work page

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.