pith. sign in

arxiv: 2606.04522 · v1 · pith:MNKNU45Mnew · submitted 2026-06-03 · 💻 cs.IR · cs.AI· cs.DB· cs.LG

ANN Search: Recall What Matters

Dimitris Dimitropoulos , Nikos Mamoulis This is my paper

Pith reviewed 2026-06-28 04:26 UTC · model grok-4.3

classification 💻 cs.IR cs.AIcs.DBcs.LG
keywords approximate nearest neighborANN searchRecall@kapproximation ratioinformation retrievalretrieval-augmented generationevaluation metricsdownstream task performance
0
0 comments X

The pith

Replacing Recall@k with 1/Ratio@k lets ANN search reach quality thresholds at lower cost while tracking downstream utility more closely.

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

The paper claims that Recall@k, the fraction of exact neighbors retrieved, overstates the true cost of approximation in ANN search and does not best reflect result quality. It proposes 1/Ratio@k, the inverse of the average ratio between distances to retrieved neighbors and true neighbors, as a judge-free alternative that depends only on standard benchmark inputs. Experiments across datasets with varying intrinsic dimensionalities show that tuning algorithms for 1/Ratio@k meets operational quality levels with substantially less computation than Recall@k requires. Downstream indicators such as classification precision, semantic similarity, BERTScore, and LLM judgments stay stable even when Recall@k falls, while 1/Ratio@k follows those indicators more tightly. The result is that current Recall@k-based practices impose unnecessary overhead in ANN systems.

Core claim

The paper establishes that 1/Ratio@k is a more accurate proxy for ANN quality than Recall@k because it measures distance approximation directly rather than set overlap. Across diverse datasets, optimizing for 1/Ratio@k reaches usable quality thresholds at lower computational cost, and downstream task performance remains stable even as Recall@k declines, with 1/Ratio@k tracking true utility more reliably.

What carries the argument

The inverse approximation ratio 1/Ratio@k, the reciprocal of the mean ratio of distances to the k retrieved neighbors versus the k true neighbors.

If this is right

  • ANN algorithms can be tuned to lower computational budgets while still meeting operational quality thresholds.
  • Downstream performance in classification and retrieval-augmented generation remains stable despite reduced exact neighbor overlap.
  • 1/Ratio@k correlates more closely with task-specific quality signals than Recall@k across efficiency and utility axes.
  • Standard ANN benchmarks can adopt 1/Ratio@k to avoid overstating the cost of approximation.

Where Pith is reading between the lines

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

  • Production ANN deployments could reduce compute by monitoring 1/Ratio@k directly instead of targeting high Recall@k.
  • Index designs might shift toward tolerating set mismatches when distances remain accurate.
  • The metric could extend to other approximate retrieval settings where fidelity of results matters more than exact matches.
  • Evaluation pipelines could become more parameter-free by relying on 1/Ratio@k computed from existing benchmark data.

Load-bearing premise

The approximation ratio between retrieved and true neighbor distances predicts downstream task quality more reliably than exact neighbor overlap across the tested datasets and tasks.

What would settle it

A dataset or downstream task in which 1/Ratio@k stays high yet classification accuracy, semantic similarity, or LLM-graded quality drops sharply while Recall@k is low.

Figures

Figures reproduced from arXiv: 2606.04522 by Dimitris Dimitropoulos, Nikos Mamoulis.

Figure 1
Figure 1. Figure 1: Approximate and exact 𝑘NN results in a RAG down￾stream task. The embeddings missed by the ANN algorithm are not necessarily of much higher quality compared to the retrieved ones not in the exact 𝑘NN set. Recall@𝑘 gives un￾reasonably high penalty to retrieval quality. Other alternatives to Recall@𝑘. Several recent works have ques￾tioned the effectiveness of Recall@𝑘 as a metric for ANN evaluation [10, 33, 5… view at source ↗
Figure 2
Figure 2. Figure 2: Maximum achievable QPS under Recall@100 and 1/Ratio@100 across target quality thresholds [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Minimum distance computations per query to reach quality threshold [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: 6.2 Classification (RQ4) The cost-accuracy experiments show that 1/Ratio@𝑘 reaches oper￾ational thresholds at substantially lower cost than Recall@𝑘. But a metric with easy-to-achieve quality targets, like 1/Ratio@𝑘, is only useful if it actually tracks what matters: downstream task quality. In this subsection we test this directly by investigating whether 1/Ratio@𝑘 is aligned with downstream task retrieva… view at source ↗
Figure 4
Figure 4. Figure 4: QPS speedup of 1/Ratio over Recall as a function of [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Average change in Recall@𝑘 and 1/Ratio@𝑘 relative to their 𝑘=1 values, as 𝑘 grows from 1 to 100, across all parameter configurations per algorithm. Recall degrades substantially with 𝑘, while 1/Ratio remains nearly flat. 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Recall@100 40 60 80 100 % of r=1.0 Performance MNIST 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Recall@100 Fashion-MNIST 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Recall@100 CIFAR-10 0.4 0.… view at source ↗
Figure 6
Figure 6. Figure 6: LP@100 and 1/Ratio@100 (normalized to exact search) versus synthetic recall. Both metrics stay near 1.0 as recall [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Mean Absolute Deviation (MAD) of Recall and [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Downstream generation quality (Semantic Similarity, BERTScore F1, Grade/10) and geometric stability ( [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: MAD of Recall and 1/Ratio from true RAG quality [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
read the original abstract

Approximate nearest neighbor (ANN) search has become a core primitive in information retrieval and modern machine learning tasks, from classification to retrieval-augmented generation. The community evaluates and tunes ANN algorithms primarily on their throughput at a given Recall@k, the fraction of true exact neighbors retrieved. We argue that what really matters in ANN search is the quality of the retrieved results and not their overlap with the true kNN set. We show that using Recall@k to assess retrieval quality forces unnecessary computational overhead and investigate replacing it by 1/Ratio@k, the inverse approximation ratio. 1/Ratio@k evaluates the differences between the distances of the retrieved and true neighbors. It is judge-free, hyperparameter-free, and computable from standard ANN benchmark inputs alone. We benchmark state-of-the-art ANN algorithms across diverse datasets spanning a wide range of intrinsic dimensionalities, evaluating the two metrics comprehensively across efficiency, downstream classification, and retrieval-augmented generation. On the efficiency axis, optimizing for 1/Ratio@k reaches operational quality thresholds at a substantially lower computational cost than Recall@k. In downstream tasks, performance indicators (label precision, semantic similarity, BERTScore, and LLM-graded quality) remain highly stable even when Recall@k drops significantly. The inverse approximation ratio, on the other hand, closely mirrors this stability, tracking true utility much better than Recall@k. Ultimately, while Recall@k overstates the true cost of approximation, 1/Ratio@k offers a more accurate, deployable proxy for actual ANN quality.

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.

Referee Report

2 major / 3 minor

Summary. The paper argues that Recall@k, the standard metric for ANN search quality, overstates the computational cost of approximation by focusing on exact neighbor overlap rather than result quality. It proposes replacing it with 1/Ratio@k (the inverse approximation ratio based on distances to true vs. retrieved neighbors), which is judge-free, hyperparameter-free, and computable from standard ANN benchmark inputs. Through benchmarks on diverse datasets spanning intrinsic dimensionalities, the authors show that optimizing for 1/Ratio@k achieves operational quality at lower cost than Recall@k, while downstream task metrics (classification precision, semantic similarity, BERTScore, LLM judgments) remain stable even as Recall@k drops, with 1/Ratio@k tracking utility more closely.

Significance. If the empirical correlations hold, this work could shift ANN evaluation practices away from overlap-based metrics toward distance-ratio proxies that better reflect real utility in IR and ML pipelines, enabling more efficient algorithm tuning without sacrificing downstream performance. Strengths include the metric's direct computability from existing benchmark data, comprehensive evaluation across efficiency and multiple downstream tasks, and absence of post-hoc parameter tuning or invented entities in the core argument.

major comments (2)
  1. [§4] §4 (efficiency benchmarks): the claim that 1/Ratio@k reaches 'operational quality thresholds' at substantially lower cost requires explicit definition of those thresholds (e.g., specific downstream metric cutoffs) and confirmation that they were fixed a priori rather than chosen to favor the new metric; without this, the cost comparison risks being circular with the stability results.
  2. [Table 2] Table 2 / downstream task results: the reported stability of label precision and BERTScore when Recall@k varies but 1/Ratio@k is controlled needs per-dataset statistical tests (e.g., correlation coefficients with p-values) to establish that the tracking advantage is not driven by a subset of low-dimensionality datasets.
minor comments (3)
  1. [Abstract / §2] The abstract states the metric is 'hyperparameter-free,' but the definition of Ratio@k should include an explicit equation (likely in §2) showing it uses only the distances already present in standard ANN inputs with no tunable parameters.
  2. [Figures] Figure captions for efficiency curves should clarify the exact k values and dataset intrinsic dimensionalities used, to allow direct reproduction from the reported numbers.
  3. [§1] A brief related-work paragraph contrasting 1/Ratio@k with existing approximation-ratio measures in ANN literature (e.g., those in the original ANN benchmarks) would strengthen the novelty claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. We address each major comment below.

read point-by-point responses

  1. Referee: [§4] §4 (efficiency benchmarks): the claim that 1/Ratio@k reaches 'operational quality thresholds' at substantially lower cost requires explicit definition of those thresholds (e.g., specific downstream metric cutoffs) and confirmation that they were fixed a priori rather than chosen to favor the new metric; without this, the cost comparison risks being circular with the stability results.

    Authors: We agree that the operational quality thresholds must be defined explicitly. In the revised manuscript we will add a precise definition in §4: operational quality is reached when a downstream metric attains at least 90 % of its exact-search value, with the 90 % cutoff selected from prior IR literature on acceptable approximation loss before any efficiency curves were examined. We will also state that the cost comparisons were performed after fixing this cutoff, thereby removing any circular dependence on the stability plots. revision: yes

  2. Referee: [Table 2] Table 2 / downstream task results: the reported stability of label precision and BERTScore when Recall@k varies but 1/Ratio@k is controlled needs per-dataset statistical tests (e.g., correlation coefficients with p-values) to establish that the tracking advantage is not driven by a subset of low-dimensionality datasets.

    Authors: We accept the request for statistical support. The revision will augment the Table 2 analysis with per-dataset Pearson correlation coefficients and p-values between each candidate metric (Recall@k and 1/Ratio@k) and every downstream indicator. These tests will be reported for all datasets, confirming that the tracking advantage of 1/Ratio@k is not confined to low-dimensionality subsets. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core argument defines 1/Ratio@k explicitly from distances already present in standard ANN benchmark inputs and validates its superiority over Recall@k via direct empirical comparisons on efficiency and downstream task stability (classification precision, BERTScore, LLM judgments) across multiple datasets. No load-bearing step reduces by construction to a self-definition, a fitted parameter renamed as a prediction, or a self-citation chain; the metric is computable without hyperparameters and the claims rest on external task metrics independent of the ratio definition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the definition of 1/Ratio@k as the inverse of the approximation ratio and on the empirical observation that downstream metrics track this quantity more closely than Recall@k. No additional free parameters, axioms beyond standard mathematics, or invented entities are introduced.

pith-pipeline@v0.9.1-grok · 5804 in / 1169 out tokens · 47083 ms · 2026-06-28T04:26:50.969513+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

67 extracted references · 38 canonical work pages · 3 internal anchors

  1. [1]

    Laurent Amsaleg and Hervé Jégou. 2026. Datasets for approximate nearest neighbor search. http://corpus-texmex.irisa.fr/. (Accessed: February, 2026)

    2026

  2. [2]

    Mount, Nathan S

    Sunil Arya, David M. Mount, Nathan S. Netanyahu, Ruth Silverman, and Angela Y. Wu. 1998. An optimal algorithm for approximate nearest neighbor searching fixed dimensions.J. ACM45, 6 (Nov. 1998), 891–923. doi:10.1145/293347.293348

    work page doi:10.1145/293347.293348 1998

  3. [3]

    Martin Aumüller, Erik Bernhardsson, and Alexander John Faithfull. 2020. ANN- Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms. Inf. Syst.87 (2020). doi:10.1016/J.IS.2019.02.006

    work page doi:10.1016/j.is.2019.02.006 2020

  4. [4]

    Martin Aumüller and Matteo Ceccarello. 2019. The Role of Local Intrinsic Dimensionality in Benchmarking Nearest Neighbor Search. InSimilarity Search and Applications - 12th International Conference, SISAP 2019, Newark, NJ, USA, October 2-4, 2019, Proceedings (Lecture Notes in Computer Science), Giuseppe Amato, Claudio Gennaro, Vincent Oria, and Milos Rado...

    work page doi:10.1007/978-3-030-32047-8_11 2019

  5. [5]

    Lempitsky

    Artem Babenko and Victor S. Lempitsky. 2012. The inverted multi-index. In2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, USA, June 16-21, 2012. IEEE Computer Society, 3069–3076. doi:10.1109/CVPR.2012. 6248038

    work page doi:10.1109/cvpr.2012 2012

  6. [6]

    Jon Louis Bentley. 1975. Multidimensional Binary Search Trees Used for Associa- tive Searching.Commun. ACM18, 9 (1975), 509–517. doi:10.1145/361002.361007

    work page doi:10.1145/361002.361007 1975

  7. [7]

    Beyer, Jonathan Goldstein, Raghu Ramakrishnan, and Uri Shaft

    Kevin S. Beyer, Jonathan Goldstein, Raghu Ramakrishnan, and Uri Shaft. 1999. When Is ”Nearest Neighbor” Meaningful?. InDatabase Theory - ICDT ’99, 7th In- ternational Conference, Jerusalem, Israel, January 10-12, 1999, Proceedings (Lecture Notes in Computer Science), Catriel Beeri and Peter Buneman (Eds.). Springer, 217–235. doi:10.1007/3-540-49257-7_15

    work page doi:10.1007/3-540-49257-7_15 1999

  8. [8]

    Manos Chatzakis, Yannis Papakonstantinou, and Themis Palpanas. 2025. DARTH: Declarative Recall Through Early Termination for Approximate Nearest Neighbor Search.Proc. ACM Manag. Data3, 4 (2025), 242:1–242:26. doi:10.1145/3749160

    work page doi:10.1145/3749160 2025

  9. [9]

    Sean Wang

    Meng Chen, Kai Zhang, Zhenying He, Yinan Jing, and X. Sean Wang. 2024. RoarGraph: A Projected Bipartite Graph for Efficient Cross-Modal Approximate Nearest Neighbor Search.Proc. VLDB Endow.17, 11 (2024), 2735–2749. doi:10. 14778/3681954.3681959

    arXiv 2024

  10. [10]

    Tingyang Chen, Cong Fu, Jiahua Wu, Haotian Wu, Hua Fan, Xiangyu Ke, Yunjun Gao, Yabo Ni, and Anxiang Zeng. 2025. Reveal Hidden Pitfalls and Navigate Next Generation of Vector Similarity Search from Task-Centric Views.CoRR abs/2512.12980 (2025). arXiv:2512.12980 doi:10.48550/ARXIV.2512.12980

    work page doi:10.48550/arxiv.2512.12980 2025

  11. [11]

    Sanjoy Dasgupta and Yoav Freund. 2008. Random projection trees and low dimensional manifolds. InProceedings of the 40th Annual ACM Symposium on Theory of Computing, Victoria, British Columbia, Canada, May 17-20, 2008, Cynthia Dwork (Ed.). ACM, 537–546. doi:10.1145/1374376.1374452

    work page doi:10.1145/1374376.1374452 2008

  12. [12]

    Mirrokni

    Mayur Datar, Nicole Immorlica, Piotr Indyk, and Vahab S. Mirrokni. 2004. Locality-sensitive hashing scheme based on p-stable distributions. InProceedings of the 20th ACM Symposium on Computational Geometry, Brooklyn, New York, USA, June 8-11, 2004, Jack Snoeyink and Jean-Daniel Boissonnat (Eds.). ACM, 253–262. doi:10.1145/997817.997857

    work page doi:10.1145/997817.997857 2004

  13. [13]

    Laxman Dhulipala, Majid Hadian, Rajesh Jayaram, Jason Lee, and Vahab Mirrokni

  14. [14]

    MUVERA: Multi-Vector Retrieval via Fixed Dimensional Encodings

    MUVERA: Multi-Vector Retrieval via Fixed Dimensional Encodings.CoRR abs/2405.19504 (2024). arXiv:2405.19504 doi:10.48550/ARXIV.2405.19504

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2405.19504 2024

  15. [16]

    Elena Facco, Maria d’Errico, Alex Rodriguez, and Alessandro Laio. 2018. Estimat- ing the intrinsic dimension of datasets by a minimal neighborhood information. CoRRabs/1803.06992 (2018). arXiv:1803.06992 http://arxiv.org/abs/1803.06992

    Pith/arXiv arXiv 2018

  16. [17]

    Cong Fu, Changxu Wang, and Deng Cai. 2022. High Dimensional Similarity Search With Satellite System Graph: Efficiency, Scalability, and Unindexed Query Compatibility.IEEE Trans. Pattern Anal. Mach. Intell.44, 8 (2022), 4139–4150. doi:10.1109/TPAMI.2021.3067706

    work page doi:10.1109/tpami.2021.3067706 2022

  17. [18]

    Cong Fu, Chao Xiang, Changxu Wang, and Deng Cai. 2019. Fast Approximate Nearest Neighbor Search With The Navigating Spreading-out Graph.Proc. VLDB Endow.12, 5 (2019), 461–474. doi:10.14778/3303753.3303754

    work page doi:10.14778/3303753.3303754 2019

  18. [19]

    Jianyang Gao, Yutong Gou, Yuexuan Xu, Yongyi Yang, Cheng Long, and Ray- mond Chi-Wing Wong. 2025. Practical and Asymptotically Optimal Quantization of High-Dimensional Vectors in Euclidean Space for Approximate Nearest Neigh- bor Search.Proc. ACM Manag. Data3, 3 (2025), 202:1–202:26. doi:10.1145/3725413

    work page doi:10.1145/3725413 2025

  19. [20]

    Jianyang Gao and Cheng Long. 2023. High-Dimensional Approximate Nearest Neighbor Search: with Reliable and Efficient Distance Comparison Operations. Proc. ACM Manag. Data1, 2 (2023), 137:1–137:27. doi:10.1145/3589282

    work page doi:10.1145/3589282 2023

  20. [21]

    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 (2024), 167. doi:10.1145/3654970

    work page doi:10.1145/3654970 2024

  21. [22]

    Tiezheng Ge, Kaiming He, Qifa Ke, and Jian Sun. 2014. Optimized Product Quantization.IEEE Trans. Pattern Anal. Mach. Intell.36, 4 (2014), 744–755. doi:10. 1109/TPAMI.2013.240

    2014

  22. [23]

    Yutong Gou, Jianyang Gao, Yuexuan Xu, and Cheng Long. 2025. SymphonyQG: Towards Symphonious Integration of Quantization and Graph for Approximate Nearest Neighbor Search.Proc. ACM Manag. Data3, 1 (2025), 80:1–80:26. doi:10. 1145/3709730

    2025

  23. [24]

    Ruiqi Guo, Philip Sun, Erik Lindgren, Quan Geng, David Simcha, Felix Chern, and Sanjiv Kumar. 2020. Accelerating Large-Scale Inference with Anisotropic Vector Quantization. InProceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event (Proceedings of Machine Learning Research). PMLR, 3887–3896. http://pr...

    2020

  24. [25]

    Piotr Indyk and Rajeev Motwani. 1998. Approximate Nearest Neighbors: Towards Removing the Curse of Dimensionality. InProceedings of the Thirtieth Annual ACM Symposium on the Theory of Computing, Dallas, Texas, USA, May 23-26, 1998, Jeffrey Scott Vitter (Ed.). ACM, 604–613. doi:10.1145/276698.276876

    work page doi:10.1145/276698.276876 1998

  25. [26]

    Elias Jääsaari, Ville Hyvönen, Matteo Ceccarello, Teemu Roos, and Martin Aumüller. 2025. VIBE: Vector Index Benchmark for Embeddings.CoRR abs/2505.17810 (2025). arXiv:2505.17810 doi:10.48550/ARXIV.2505.17810

    work page doi:10.48550/arxiv.2505.17810 2025

  26. [27]

    Suhas Jayaram Subramanya, Fnu Devvrit, Harsha Vardhan Simhadri, Ravishankar Krishnawamy, and Rohan Kadekodi. 2019. DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node. InAdvances in Neural Information Processing Systems, H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett (Eds.), Vol. 32. Curran Asso...

    2019

  27. [28]

    Hervé Jégou, Matthijs Douze, and Cordelia Schmid. 2011. Product Quantization for Nearest Neighbor Search.IEEE Trans. Pattern Anal. Mach. Intell.33, 1 (2011), 117–128. doi:10.1109/TPAMI.2010.57

    work page doi:10.1109/tpami.2010.57 2011

  28. [29]

    Cohen, and Xinghua Lu

    Qiao Jin, Bhuwan Dhingra, Zhengping Liu, William W. Cohen, and Xinghua Lu

  29. [30]

    PubMedQA: A Dataset for Biomedical Research Question Answering. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Process- ing, EMNLP-IJCNLP 2019, Hong Kong, China, November 3-7, 2019, Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan (Eds.). Associa...

    work page doi:10.18653/v1/d19-1259 2019

  30. [31]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2021. Billion-Scale Similarity Search with GPUs.IEEE Trans. Big Data7, 3 (2021), 535–547. doi:10.1109/ TBDATA.2019.2921572

    arXiv 2021

  31. [32]

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense Passage Retrieval for Open- Domain Question Answering. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020, Bonnie Webber, Trevor Cohn, Yulan He, and Ya...

    work page doi:10.18653/v1/2020.emnlp-main.550 2020

  32. [33]

    Alex Krizhevsky. 2009. Learning Multiple Layers of Features from Tiny Images. https://api.semanticscholar.org/CorpusID:18268744

    2009

  33. [34]

    Leonardo Kuffó, Elena Krippner, and Peter Boncz. 2025. PDX: A Data Layout for Vector Similarity Search.Proc. ACM Manag. Data3, 3 (2025), 196:1–196:26. doi:10.1145/3725333

    work page doi:10.1145/3725333 2025

  34. [35]

    Leonardo Kuffo, Ioanna Tsakalidou, Roberta Viti, Albert Angel, Jiří Iša, and Rastislav Lenhardt. 2026. Semantic Recall for Vector Search. doi:10.48550/arXiv. 2604.20417

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2026

  35. [36]

    Yann LeCun, Corinna Cortes, and CJ Burges. 2010. MNIST handwritten digit database.ATT Labs [Online]. A vailable: http://yann.lecun.com/exdb/mnist2 (2010)

    2010

  36. [37]

    Leonardo Kuffo, Elena Krippner, Peter Boncz. 2026. PDX: Public Data. https:// drive.google.com/drive/u/1/folders/1f76UCrU52N2wToGMFg9ir1MY8ZocrN34. (Accessed: February, 2026)

    2026

  37. [38]

    Alexandria Leto, Cecilia Aguerrebere, Ishwar Singh Bhati, Ted Willke, Mariano Tepper, and Vy Ai Vo. 2024. Toward Optimal Search and Retrieval for RAG.CoRR abs/2411.07396 (2024). arXiv:2411.07396 doi:10.48550/ARXIV.2411.07396

    work page doi:10.48550/arxiv.2411.07396 2024

  38. [39]

    Mocheng Li, Xiao Yan, Baotong Lu, Yue Zhang, James Cheng, and Chenhao Ma

  39. [40]

    ACM Manag

    Attribute Filtering in Approximate Nearest Neighbor Search: An In-depth Experimental Study.Proc. ACM Manag. Data3, 6 (2025), 1–26. doi:10.1145/ 3769763

    2025

  40. [41]

    Wen Li, Ying Zhang, Yifang Sun, Wei Wang, Mingjie Li, Wenjie Zhang, and Xuemin Lin. 2020. Approximate Nearest Neighbor Search on High Dimensional Data - Experiments, Analyses, and Improvement.IEEE Trans. Knowl. Data Eng. 32, 8 (2020), 1475–1488. doi:10.1109/TKDE.2019.2909204

    work page doi:10.1109/tkde.2019.2909204 2020

  41. [42]

    Xiangci Li and Jessica Ouyang. 2025. How Does Knowledge Selection Help Retrieval Augmented Generation?. InFindings of the Association for Compu- tational Linguistics: EMNLP 2025, Suzhou, China, November 4-9, 2025, Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, and Violet Peng (Eds.). Association for Computational Linguistics, 4104–4121. ht...

    2025

  42. [43]

    Jiahao Lou, Quan Yu, Shufeng Gong, Song Yu, Yanfeng Zhang, and Ge Yu. 2025. DGAI: Decoupled On-Disk Graph-Based ANN Index for Efficient Updates and Queries.CoRRabs/2510.25401 (2025). arXiv:2510.25401 doi:10.48550/ARXIV.2510. 25401

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2510 2025

  43. [44]

    Yury Malkov, Alexander Ponomarenko, Andrey Logvinov, and Vladimir Krylov

  44. [45]

    Syst.45 (2014), 61–68

    Approximate nearest neighbor algorithm based on navigable small world Dimitris Dimitropoulos and Nikos Mamoulis 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

  45. [46]

    Malkov and Dmitry A

    Yury A. Malkov and Dmitry A. Yashunin. 2020. Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs.IEEE Trans. Pattern Anal. Mach. Intell.42, 4 (2020), 824–836. doi:10.1109/TPAMI.2018. 2889473

    work page doi:10.1109/tpami.2018 2020

  46. [47]

    Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and An- drew Y Ng. 2011. Reading digits in natural images with unsupervised feature learning. (2011)

    2011

  47. [48]

    Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. MS MARCO: A Human Generated MAchine Reading COmprehension Dataset. InProceedings of the Workshop on Cognitive Computation: Integrating neural and symbolic approaches 2016 co-located with the 30th Annual Conference on Neural Information Processing Systems...

    2016

  48. [49]

    Elmore, and Michael J

    John Paparrizos, Ikraduya Edian, Chunwei Liu, Aaron J. Elmore, and Michael J. Franklin. 2022. Fast Adaptive Similarity Search through Variance-Aware Quantization. In38th IEEE International Conference on Data Engineering, ICDE 2022, Kuala Lumpur, Malaysia, May 9-12, 2022. IEEE, 2969–2983. doi:10.1109/ ICDE53745.2022.00268

    arXiv 2022

  49. [50]

    Spotify AB. 2025. spotify/annoy: Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk. https://github.com/ spotify/annoy. commit on main branch accessed on 2025-11-29

    2025

  50. [51]

    Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models. InProceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks 1, NeurIPS Datasets and Benchmarks 2021, December 2021, virtual, Joaquin Vanschoren...

    2021

  51. [52]

    Jianguo Wang, Xiaomeng Yi, Rentong Guo, Hai Jin, Peng Xu, Shengjun Li, Xi- angyu 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. InSIGMOD ’21: Internatio...

    work page doi:10.1145/3448016.3457550 2021

  52. [53]

    Jingdong Wang, Ting Zhang, Jingkuan Song, Nicu Sebe, and Heng Tao Shen

  53. [54]

    Pattern Anal

    A Survey on Learning to Hash.IEEE Trans. Pattern Anal. Mach. Intell.40, 4 (2018), 769–790. doi:10.1109/TPAMI.2017.2699960

    work page doi:10.1109/tpami.2017.2699960 2018

  54. [55]

    Mengzhao Wang, Haotian Wu, Xiangyu Ke, Yunjun Gao, Yifan Zhu, and Wenchao Zhou. 2025. Accelerating Graph Indexing for ANNS on Modern CPUs.Proc. ACM Manag. Data3, 3 (2025), 123:1–123:29. doi:10.1145/3725260

    work page doi:10.1145/3725260 2025

  55. [56]

    Mengzhao Wang, Xiaoliang Xu, Qiang Yue, and Yuxiang Wang. 2021. A Com- prehensive Survey and Experimental Comparison of Graph-Based Approxi- mate Nearest Neighbor Search.Proc. VLDB Endow.14, 11 (2021), 1964–1978. doi:10.14778/3476249.3476255

    work page doi:10.14778/3476249.3476255 2021

  56. [57]

    Ziqi Wang, Jingzhe Zhang, and Wei Hu. 2025. WoW: A Window-to-Window Incremental Index for Range-Filtering Approximate Nearest Neighbor Search. Proc. ACM Manag. Data3, 6 (2025), 1–27. doi:10.1145/3769843

    work page doi:10.1145/3769843 2025

  57. [58]

    Zikai Wang, Qianxi Zhang, Baotong Lu, Qi Chen, and Cheng Tan. 2025. To- wards Robustness: A Critique of Current Vector Database Assessments.CoRR abs/2507.00379 (2025). arXiv:2507.00379 doi:10.48550/ARXIV.2507.00379

    work page doi:10.48550/arxiv.2507.00379 2025

  58. [59]

    Roger Weber, Hans-Jörg Schek, and Stephen Blott. 1998. A Quantitative Analysis and Performance Study for Similarity-Search Methods in High-Dimensional Spaces. InVLDB’98, Proceedings of 24rd International Conference on Very Large Data Bases, August 24-27, 1998, New York City, New York, USA, Ashish Gupta, Oded Shmueli, and Jennifer Widom (Eds.). Morgan Kauf...

    1998

  59. [60]

    Jiuqi Wei, Xiaodong Lee, Zhenyu Liao, Themis Palpanas, and Botao Peng. 2025. Subspace Collision: An Efficient and Accurate Framework for High-dimensional Approximate Nearest Neighbor Search.Proc. ACM Manag. Data3, 1 (2025), 79:1–79:29. doi:10.1145/3709729

    work page doi:10.1145/3709729 2025

  60. [61]

    Jiuqi Wei, Botao Peng, Xiaodong Lee, and Themis Palpanas. 2024. DET-LSH: A Locality-Sensitive Hashing Scheme with Dynamic Encoding Tree for Approx- imate Nearest Neighbor Search.Proc. VLDB Endow.17, 9 (2024), 2241–2254. doi:10.14778/3665844.3665854

    work page doi:10.14778/3665844.3665854 2024

  61. [62]

    Han Xiao, Kashif Rasul, and Roland Vollgraf. 2017. Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms.CoRRabs/1708.07747 (2017). arXiv:1708.07747 http://arxiv.org/abs/1708.07747

    Pith/arXiv arXiv 2017

  62. [63]

    Yahoo Japan. 2018. NGT: Neighborhood Graph and Tree for Indexing High- dimensional Data. https://github.com/yahoojapan/NGT. Accessed: 2026-05-20

    2018

  63. [64]

    Donghui Yan, Yingjie Wang, Jin Wang, Honggang Wang, and Zhenpeng Li. 2018. K-nearest Neighbor Search by Random Projection Forests. InIEEE International Conference on Big Data (IEEE BigData 2018), Seattle, W A, USA, December 10-13, 2018, Naoki Abe, Huan Liu, Calton Pu, Xiaohua Hu, Nesreen K. Ahmed, Mu Qiao, Yang Song, Donald Kossmann, Bing Liu, Kisung Lee,...

    work page doi:10.1109/bigdata.2018.8622307 2018

  64. [65]

    Cohen, Ruslan Salakhutdinov, and Christopher D

    Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W. Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering. InConference on Empirical Methods in Natural Language Processing (EMNLP)

    2018

  65. [66]

    Qiang Yue, Xiaoliang Xu, Yuxiang Wang, Yikun Tao, and Xuliyuan Luo. 2024. Routing-Guided Learned Product Quantization for Graph-Based Approximate Nearest Neighbor Search. In40th IEEE International Conference on Data Engi- neering, ICDE 2024, Utrecht, The Netherlands, May 13-16, 2024. IEEE, 4870–4883. doi:10.1109/ICDE60146.2024.00370

    work page doi:10.1109/icde60146.2024.00370 2024

  66. [67]

    Weinberger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi

  67. [68]

    https://api.semanticscholar.org/CorpusID:127986044

    BERTScore: Evaluating Text Generation with BERT.ArXivabs/1904.09675 (2019). https://api.semanticscholar.org/CorpusID:127986044

    Pith/arXiv arXiv 1904