Pith. sign in

REVIEW 3 major objections 1 minor 2 cited by

A procedural generator can create billions of algorithm-discovery tasks so agents can be trained and evaluated without the contamination and saturation of fixed suites.

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 ↗

T0 review · grok-4.5

2026-07-13 22:54 UTC pith:LC6CWMAY

load-bearing objection We do not have the DiscoGen paper—only its abstract; the supplied full text is an unrelated iris-PAD manuscript, so the central claims cannot be checked. the 3 major comments →

arxiv 2603.17863 v2 pith:LC6CWMAY submitted 2026-03-18 cs.LG cs.AI

DiscoGen: Procedural Generation of Algorithm Discovery Tasks in Machine Learning

Alexander D. Goldie , Zilin Wang , Adrian Hayler , Deepak Nathani , Edan Toledo , Ken Thampiratwong , Aleksandra Kalisz , Michael Beukman
show 12 more authors
Hannah Erlebach Alistair Letcher Shashank Reddy Clarisse Wibault Theo Wolf Charles O'Neill Uljad Berdica Nicholas Roberts Saeed Rahmani Roberta Raileanu Shimon Whiteson Jakob N. Foerster
This is my paper
classification cs.LG cs.AI
keywords algorithm discoveryprocedural generationDiscoGenDiscoBenchalgorithm discovery agentsloss functionsoptimisersmachine learning
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.

Existing suites for testing systems that invent machine-learning algorithms are limited by weak evaluation methods, data contamination, and problems that are either solved or too alike. DiscoGen addresses this by procedurally generating algorithm-discovery tasks—such as inventing optimisers for reinforcement learning or loss functions for image classification—from a handful of configuration parameters. The generator produces billions of tasks that vary in difficulty and complexity across multiple machine-learning domains. A fixed subset called DiscoBench supplies a stable evaluation benchmark, while the open-ended generator itself can be used to optimise the discovery agents. The work shows this pipeline in action with scaling experiments on automated prompt tuning and sketches further research directions that the generator makes possible.

Core claim

DiscoGen shows that a small set of configuration parameters is enough to procedurally generate a vast, open-ended space of algorithm-discovery tasks whose difficulty and domain coverage let researchers both train algorithm-discovery agents and evaluate them on a clean, contamination-free fixed subset (DiscoBench).

What carries the argument

DiscoGen: a procedural task generator that maps a compact configuration space onto billions of concrete algorithm-discovery problems (optimiser design, loss-function design, etc.) of controlled difficulty.

Load-bearing premise

A small number of configuration knobs can generate tasks whose diversity and scientific value really span useful algorithm discovery, rather than producing superficial variants that agents can game without inventing transferable algorithms.

What would settle it

Train an agent solely on DiscoGen-generated tasks, then test whether the algorithms it invents improve performance on held-out, real-world algorithm-design problems that were never encoded in the generator; failure of transfer would falsify the claim that the procedural tasks are scientifically useful.

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

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 / 1 minor

Summary. The manuscript claims to introduce DiscoGen, a procedural generator of algorithm-discovery tasks for machine learning (e.g., optimisers for RL, loss functions for image classification). From a small set of configuration parameters it is said to produce billions of tasks of varying difficulty across ML fields, thereby addressing contamination, saturation and weak evaluation methodologies of prior suites. A fixed subset called DiscoBench is proposed for principled evaluation of algorithm discovery agents (ADAs), and scaling experiments on automated prompt tuning are cited as a demonstration of utility. The abstract further asserts open-source release and lists ambitious research directions. The supplied full-text body, however, is an unrelated paper (VISER) on open-set iris presentation-attack detection that contains none of the claimed generator, tasks, metrics or experiments.

Significance. If the claims were substantiated, a large, contamination-resistant procedural suite for algorithm discovery would be a genuine contribution to AutoML and automated scientific discovery, enabling systematic optimisation and evaluation of ADAs. The abstract’s emphasis on open-source release and a fixed evaluation subset (DiscoBench) would further strengthen reproducibility. Because the body of the document is an entirely different manuscript, none of these contributions can be assessed or credited from the supplied text.

major comments (3)
  1. The full manuscript text supplied under the DiscoGen identifier is the VISER iris-PAD paper (arXiv:2603.17859). It contains no description of DiscoGen, no configuration parameters, no task formalisms, no DiscoBench construction, and no scaling experiments on prompt tuning. Consequently every load-bearing claim of the abstract is unsupported by the document under review.
  2. Without any technical section defining the generator, it is impossible to verify the central assertion that a small number of configuration parameters yields billions of scientifically useful tasks of varying difficulty rather than superficial variants that agents can game. The abstract’s claim therefore cannot be evaluated for soundness.
  3. The promised demonstration of utility (scaling experiments for automated prompt tuning) and the proposed research directions are likewise absent from the body. No tables, figures or results corresponding to DiscoGen appear, rendering the empirical contribution unverifiable.
minor comments (1)
  1. Metadata and abstract refer to DiscoGen / DiscoBench while the body, figures, tables and references belong to an unrelated iris-PAD study; this mismatch must be corrected before any scientific review is possible.

Circularity Check

0 steps flagged

No circularity: DiscoGen abstract presents an independent generator; supplied body is an unrelated iris-PAD paper with no derivation chain to inspect.

full rationale

The claimed paper (DiscoGen) supplies only an abstract that introduces a procedural task generator as an independent artifact whose outputs can train or evaluate algorithm-discovery agents; no equations, fitted parameters, uniqueness theorems, or self-referential definitions appear. The full manuscript text provided is an entirely different work (VISER, open-set iris presentation-attack detection) whose experimental comparisons of saliency maps, DenseNet baselines, and foundation-model embeddings contain no theoretical derivation, no fitted constants renamed as predictions, and no load-bearing self-citation chains of the kinds listed. Consequently no circular step can be exhibited by quotation and reduction. The abstract’s generator is self-contained against external benchmarks (open-source release, fixed DiscoBench subset) and exhibits none of the six circularity patterns. Score 0 is therefore required.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 3 invented entities

Abstract-only review. The load-bearing premises are that (1) existing algorithm-discovery suites suffer from contamination, saturation and poor evaluation, (2) a small configuration space can generate scientifically meaningful and diverse tasks at billion scale, and (3) performance of ADAs on the generated tasks (and on DiscoBench) is a valid proxy for real algorithm-invention ability. No free parameters or invented physical entities appear; the main invented software entities are DiscoGen itself and DiscoBench.

axioms (3)
  • domain assumption Existing algorithm-discovery task suites suffer from poor evaluation methodologies, data contamination, and saturated or very similar problems.
    Stated as motivation in the abstract; treated as background fact without quantitative evidence in the abstract.
  • ad hoc to paper A small number of configuration parameters is sufficient to generate billions of tasks of varying difficulty and complexity that are useful for optimising and evaluating algorithm discovery agents.
    Core design claim of DiscoGen; not derived from prior theory in the abstract.
  • domain assumption Success of procedural generation in reinforcement learning transfers to the domain of algorithm-discovery tasks.
    Explicitly cited as motivation; analogy rather than proof.
invented entities (3)
  • DiscoGen independent evidence
    purpose: Procedural generator of algorithm-discovery tasks for machine learning.
    Central software artifact introduced by the paper; independent evidence would be the open-source release and external use.
  • DiscoBench no independent evidence
    purpose: Fixed, small subset of DiscoGen tasks for principled evaluation of ADAs.
    Evaluation suite defined by the authors; independent evidence would be community adoption and external re-runs.
  • Algorithm Discovery Agents (ADAs) no independent evidence
    purpose: Agents that invent ML algorithms and that can be optimised/evaluated on DiscoGen tasks.
    Terminological framing of the systems under study; not a new physical entity.

pith-pipeline@v1.1.0-grok45 · 17019 in / 2744 out tokens · 30910 ms · 2026-07-13T22:54:29.214006+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of DiscoGen: Procedural Generation of Algorithm Discovery Tasks in Machine Learning." pith.science (2026). https://pith.science/paper/LC6CWMAY

@misc{pith2026260317863,
  author       = {Pith},
  title        = {Pith review of: DiscoGen: Procedural Generation of Algorithm Discovery Tasks in Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LC6CWMAY}},
  note         = {Machine review of arXiv:2603.17863}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Automating the development of machine learning algorithms has the potential to unlock new breakthroughs. However, our ability to improve and evaluate algorithm discovery systems has thus far been limited by existing task suites. They suffer from many issues, such as: poor evaluation methodologies; data contamination; and containing saturated or very similar problems. Here, we introduce DiscoGen, a procedural generator of algorithm discovery tasks for machine learning, such as developing optimisers for reinforcement learning or loss functions for image classification. Motivated by the success of procedural generation in reinforcement learning, DiscoGen spans billions of tasks of varying difficulty and complexity from a range of machine learning fields. These tasks are specified by a small number of configuration parameters and can be used to optimise algorithm discovery agents (ADAs). We present DiscoBench, a fixed, small subset of DiscoGen tasks for principled evaluation of ADAs. Finally, we propose a number of ambitious, impactful research directions enabled by DiscoGen, and demonstrate its use for ADA optimisation through scaling experiments for automated prompt tuning. DiscoGen is released open-source at https://github.com/AlexGoldie/discogen.

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. Heuresis: Search Strategies for Autonomous AI Research Agents Across Quality, Diversity and Novelty

    cs.AI 2026-06 accept novelty 6.0

    Heuresis evaluates six search strategies for LLM research agents and shows they steer ideas along quality-diversity-novelty axes but fail to generate novel ideas that match or exceed known high-performing recipes.

  2. Heuresis: Search Strategies for Autonomous AI Research Agents Across Quality, Diversity and Novelty

    cs.AI 2026-06 unverdicted novelty 6.0

    Heuresis evaluates six search strategies for autonomous ML research agents and finds that novel ideas are rare, none rated original, and only one reaches top-10 quality while strategies steer axes but do not expand th...

Reference graph

Works this paper leans on

71 extracted references · 5 linked inside Pith · cited by 1 Pith paper

  1. [1]

    com / CVRL/OpenSourceIrisRecognition/

    University of Notre Dame Open Source Iris Recog- nition Repository.https : / / github . com / CVRL/OpenSourceIrisRecognition/. Ac- cessed: 12-14-25. 4

  2. [2]

    Accessed: 03-12-2021

    Chinese Academy of Sciences Institute of Automation (CASIA) Datasets Webpage. Accessed: 03-12-2021. 3

    2021

  3. [3]

    http://zbum.ia.pw.edu.pl/EN/node/46,

    Warsaw University of Technology Datasets Webpage. http://zbum.ia.pw.edu.pl/EN/node/46,

  4. [4]

    Iris presentation attack detection: Where are we now?Pattern Recognition Letters, 138: 483–489, 2020

    Aidan Boyd, Zhaoyuan Fang, Adam Czajka, and Kevin W Bowyer. Iris presentation attack detection: Where are we now?Pattern Recognition Letters, 138: 483–489, 2020. 2 9

    2020

  5. [5]

    Bowyer, and Adam Czajka

    Aidan Boyd, Kevin W. Bowyer, and Adam Czajka. Human-aided saliency maps improve generalization of deep learning. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vi- sion (WACV), pages 2735–2744, 2022. 3, 4

    2022

  6. [6]

    Comprehen- sive study in open-set iris presentation attack detec- tion.IEEE Transactions on Information Forensics and Security, 18:3238–3250, 2023

    Aidan Boyd, Jeremy Speth, Lucas Parzianello, Kevin W Bowyer, and Adam Czajka. Comprehen- sive study in open-set iris presentation attack detec- tion.IEEE Transactions on Information Forensics and Security, 18:3238–3250, 2023. 2, 8

    2023

  7. [7]

    Cyborg: Blending human saliency into the loss improves deep learning-based synthetic face detection

    Aidan Boyd, Patrick Tinsley, Kevin W Bowyer, and Adam Czajka. Cyborg: Blending human saliency into the loss improves deep learning-based synthetic face detection. InProceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision, pages 6108–6117, 2023. 2, 3

    2023

  8. [8]

    Density-based clustering based on hierarchi- cal density estimates

    Ricardo JGB Campello, Davoud Moulavi, and J ¨org Sander. Density-based clustering based on hierarchi- cal density estimates. InPacific-Asia conference on knowledge discovery and data mining, pages 160–

  9. [9]

    Support-vector networks.Machine learning, 20(3):273–297, 1995

    Corinna Cortes and Vladimir Vapnik. Support-vector networks.Machine learning, 20(3):273–297, 1995. 4

    1995

  10. [10]

    The regression analysis of binary se- quences.Journal of the Royal Statistical Society Se- ries B: Statistical Methodology, 20(2):215–232, 1958

    David R Cox. The regression analysis of binary se- quences.Journal of the Royal Statistical Society Se- ries B: Statistical Methodology, 20(2):215–232, 1958. 4

    1958

  11. [11]

    Mentor: Human perception-guided pretraining for increased general- ization

    Colton R Crum and Adam Czajka. Mentor: Human perception-guided pretraining for increased general- ization. In2025 IEEE/CVF Winter Conference on Ap- plications of Computer Vision (WACV), pages 7470–

  12. [12]

    Teaching ai to teach: Leveraging lim- ited human salience data into unlimited saliency-based training.arXiv preprint arXiv:2306.05527, 2023

    Colton R Crum, Aidan Boyd, Kevin Bowyer, and Adam Czajka. Teaching ai to teach: Leveraging lim- ited human salience data into unlimited saliency-based training.arXiv preprint arXiv:2306.05527, 2023. 2

    Pith/arXiv arXiv 2023

  13. [13]

    Presentation at- tack detection for iris recognition: An assessment of the state-of-the-art.ACM Computing Surveys (CSUR), 51(4):1–35, 2018

    Adam Czajka and Kevin W Bowyer. Presentation at- tack detection for iris recognition: An assessment of the state-of-the-art.ACM Computing Surveys (CSUR), 51(4):1–35, 2018. 2

    2018

  14. [14]

    On the detection of digital face ma- nipulation

    Hao Dang, Feng Liu, Joel Stehouwer, Xiaoming Liu, and Anil K Jain. On the detection of digital face ma- nipulation. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern recognition, pages 5781–5790, 2020. 2

    2020

  15. [15]

    Human attention in visual question answering: Do humans and deep networks look at the same regions?Computer Vision and Image Understanding, 163:90–100, 2017

    Abhishek Das, Harsh Agrawal, Larry Zitnick, Devi Parikh, and Dhruv Batra. Human attention in visual question answering: Do humans and deep networks look at the same regions?Computer Vision and Image Understanding, 163:90–100, 2017. 3

    2017

  16. [16]

    How iris recognition works

    John Daugman. How iris recognition works. InThe essential guide to image processing, pages 715–739. Elsevier, 2009. 2

    2009

  17. [17]

    Autosight: Automatic eye tracking-based system for immediate grading of human expertise

    Byron Dowling, Jozef Porubcin, and Adam Czajka. Autosight: Automatic eye tracking-based system for immediate grading of human expertise. In2025 IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), pages 355–366, 2025. 3, 9

    2025

  18. [18]

    the best of both worlds!

    Sukru Eraslan, Yeliz Yesilada, and Simon Harper. “the best of both worlds!” integration of web page and eye tracking data driven approaches for automatic aoi de- tection.ACM Transactions on the Web (TWEB), 14 (1):1–31, 2020. 4

    2020

  19. [19]

    A density-based algorithm for dis- covering clusters in large spatial databases with noise

    Martin Ester, Hans-Peter Kriegel, J ¨org Sander, Xi- aowei Xu, et al. A density-based algorithm for dis- covering clusters in large spatial databases with noise. Inkdd, pages 226–231, 1996. 4

    1996

  20. [20]

    Iris liveness detection based on quality related features

    Javier Galbally, Jaime Ortiz-Lopez, Julian Fierrez, and Javier Ortega-Garcia. Iris liveness detection based on quality related features. In2012 5th IAPR Inter- national Conference on Biometrics (ICB), pages 271–

  21. [21]

    Measuring human perception to improve handwritten document transcription.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 44(10):6594–6601, 2021

    Samuel Grieggs, Bingyu Shen, Greta Rauch, Pei Li, Jiaqi Ma, David Chiang, Brian Price, and Walter J Scheirer. Measuring human perception to improve handwritten document transcription.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 44(10):6594–6601, 2021. 2

    2021

  22. [22]

    In artificial intel- ligence (ai) we trust: A qualitative investigation of ai technology acceptance.Journal of Business Logistics, 43(3):388–412, 2022

    Abhinav Hasija and Terry L Esper. In artificial intel- ligence (ai) we trust: A qualitative investigation of ai technology acceptance.Journal of Business Logistics, 43(3):388–412, 2022. 2

    2022

  23. [23]

    Human attention in image captioning: Dataset and analysis

    Sen He, Hamed R Tavakoli, Ali Borji, and Nico- las Pugeault. Human attention in image captioning: Dataset and analysis. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 8529–8538, 2019. 2

    2019

  24. [24]

    Human gaze control during real- world scene perception.Trends in cognitive sciences, 7(11):498–504, 2003

    John M Henderson. Human gaze control during real- world scene perception.Trends in cognitive sciences, 7(11):498–504, 2003. 3

    2003

  25. [25]

    Densely connected convo- lutional networks

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convo- lutional networks. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 4700–4708, 2017. 3

    2017

  26. [26]

    User see, user point: gaze and cursor alignment in web search

    Jeff Huang, Ryen White, and Georg Buscher. User see, user point: gaze and cursor alignment in web search. InProceedings of the sigchi conference on hu- man factors in computing systems, pages 1341–1350,

  27. [27]

    Be specific, be clear: Bridging machine and human cap- tions by scene-guided transformer

    Yupan Huang, Zhaoyang Zeng, and Yutong Lu. Be specific, be clear: Bridging machine and human cap- tions by scene-guided transformer. InProceedings of the 2021 Workshop on Multi-Modal Pre-Training for Multimedia Understanding, pages 4–13, 2021. 2 10

    2021

  28. [28]

    Information technology — biometric presentation attack detection — part 3: Test- ing and reporting

    ISO/IEC 30107-3:2017. Information technology — biometric presentation attack detection — part 3: Test- ing and reporting. International standard, International Organization for Standardization, 2017. 3

    2017

  29. [29]

    Salicon: Saliency in context

    Ming Jiang, Shaohuang Huang, Jibin Duan, and Qi Zhao. Salicon: Saliency in context. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1072–1080, 2015. 3

    2015

  30. [30]

    A theory of reading: From eye fixations to comprehension.Psy- chological review, 87(4):329, 1980

    Marcel A Just and Patricia A Carpenter. A theory of reading: From eye fixations to comprehension.Psy- chological review, 87(4):329, 1980. 3

    1980

  31. [31]

    Analyzing and improving the image quality of stylegan

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hell- sten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of stylegan. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8110–8119, 2020. 2

    2020

  32. [32]

    Alias-free generative adversarial networks.Ad- vances in neural information processing systems, 34: 852–863, 2021

    Tero Karras, Miika Aittala, Samuli Laine, Erik H¨ark¨onen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Alias-free generative adversarial networks.Ad- vances in neural information processing systems, 34: 852–863, 2021. 2

    2021

  33. [33]

    Trustworthy artificial intelligence: a review.ACM computing surveys (CSUR), 55(2):1– 38, 2022

    Davinder Kaur, Suleyman Uslu, Kaley J Rittichier, and Arjan Durresi. Trustworthy artificial intelligence: a review.ACM computing surveys (CSUR), 55(2):1– 38, 2022. 2

    2022

  34. [34]

    Revisiting iris recognition with color cosmetic contact lenses

    Naman Kohli, Daksha Yadav, Mayank Vatsa, and Richa Singh. Revisiting iris recognition with color cosmetic contact lenses. In2013 International Con- ference on Biometrics (ICB), pages 1–7. IEEE, 2013. 3

    2013

  35. [35]

    Detecting medley of iris spoofing attacks using DESIST

    Naman Kohli, Daksha Yadav, Mayank Vatsa, Richa Singh, and Afzel Noore. Detecting medley of iris spoofing attacks using DESIST. In2016 IEEE 8th In- ternational Conference on Biometrics Theory, Appli- cations and Systems (BTAS), pages 1–6. IEEE, 2016. 3

    2016

  36. [36]

    Multifeature-based fake iris detection method.Optical Engineering, 46 (12):127204–127204, 2007

    Sung Joo Lee, Kang Ryoung Park, Youn Joo Lee, Kwanghyuk Bae, and Jaihie Kim. Multifeature-based fake iris detection method.Optical Engineering, 46 (12):127204–127204, 2007. 3

    2007

  37. [37]

    Trustworthy ai: From principles to practices.ACM Computing Sur- veys, 55(9):1–46, 2023

    Bo Li, Peng Qi, Bo Liu, Shuai Di, Jingen Liu, Jiquan Pei, Jinfeng Yi, and Bowen Zhou. Trustworthy ai: From principles to practices.ACM Computing Sur- veys, 55(9):1–46, 2023. 2

    2023

  38. [38]

    Learning what and where to attend with humans in the loop

    Drew Linsley, Dan Shiebler, Sven Eberhardt, and Thomas Serre. Learning what and where to attend with humans in the loop. InInternational Conference on Learning Representations, 2019. 2

    2019

  39. [39]

    The role of fixational eye movements in visual perception.Nature reviews neuroscience, 5 (3):229–240, 2004

    Susana Martinez-Conde, Stephen L Macknik, and David H Hubel. The role of fixational eye movements in visual perception.Nature reviews neuroscience, 5 (3):229–240, 2004. 4

    2004

  40. [40]

    Iris liveness detection com- petition (livdet-iris)–the 2025 edition

    Mahsa Mitcheff, Afzal Hossain, Samuel Webster, Sia- mul Khan, Katarzyna Roszczewska, Juan E Tapia, Fabian Stockhardt, L´azaro J Gonz´alez-Soler, Ji-Young Lim, Mirko Pollok, et al. Iris liveness detection com- petition (livdet-iris)–the 2025 edition. In2025 IEEE International Joint Conference on Biometrics (IJCB), pages 1–10. IEEE, 2025. 3

    2025

  41. [41]

    Performance of hu- mans in iris recognition: The impact of iris condition and annotation-driven verification

    Daniel Moreira, Mateusz Trokielewicz, Adam Czajka, Kevin Bowyer, and Patrick Flynn. Performance of hu- mans in iris recognition: The impact of iris condition and annotation-driven verification. In2019 IEEE Win- ter Conference on Applications of Computer Vision (WACV), pages 941–949. IEEE, 2019. 2

    2019

  42. [42]

    Deep learning for iris recognition: A sur- vey.ACM Comput

    Kien Nguyen, Hugo Proenc ¸a, and Fernando Alonso- Fernandez. Deep learning for iris recognition: A sur- vey.ACM Comput. Surv., 56(9), 2024. 2

    2024

  43. [43]

    Dinov2: Learning robust vi- sual features without supervision.arXiv preprint arXiv:2304.07193, 2023

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fer- nandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust vi- sual features without supervision.arXiv preprint arXiv:2304.07193, 2023. 3, 4

    Pith/arXiv arXiv 2023

  44. [44]

    Comparing face recog- nition algorithms to humans on challenging tasks

    Alice J O’Toole, Xaiobo An, Joseph Dunlop, Vaidehi Natu, and P Jonathon Phillips. Comparing face recog- nition algorithms to humans on challenging tasks. ACM Transactions on Applied Perception (TAP), 9(4): 1–13, 2012. 2

    2012

  45. [45]

    A parametric approach to adversarial augmentation for cross-domain iris presentation attack detection

    Debasmita Pal, Redwan Sony, and Arun Ross. A parametric approach to adversarial augmentation for cross-domain iris presentation attack detection. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 5719–5729. IEEE,

    2025

  46. [46]

    Payeye: Biometric payment and identifica- tion solutions.https://pay-eye.com/, 2025

    PayEye. Payeye: Biometric payment and identifica- tion solutions.https://pay-eye.com/, 2025. Accessed: 2026-02-09. 2

    2025

  47. [47]

    Generalist multimodal llms gain bio- metric expertise via human salience.arXiv preprint arXiv:2603.17173, 2026

    Jacob Piland, Byron Dowling, Christopher Sweet, and Adam Czajka. Generalist multimodal llms gain bio- metric expertise via human salience.arXiv preprint arXiv:2603.17173, 2026. 4

    arXiv 2026

  48. [48]

    Model focus improves performance of deep learning- based synthetic face detectors.IEEE Access, 11: 63430–63441, 2023

    Jacob C Piland, Adam Czajka, and Christopher Sweet. Model focus improves performance of deep learning- based synthetic face detectors.IEEE Access, 11: 63430–63441, 2023. 2, 4

    2023

  49. [49]

    Involuntary mo- tions of the eye during monocular fixation.Journal of experimental psychology, 40(6):687, 1950

    Floyd Ratliff and Lorrin A Riggs. Involuntary mo- tions of the eye during monocular fixation.Journal of experimental psychology, 40(6):687, 1950. 4

    1950

  50. [50]

    Visual psychophysics for making face recognition al- gorithms more explainable

    Brandon RichardWebster, So Yon Kwon, Christopher Clarizio, Samuel E Anthony, and Walter J Scheirer. Visual psychophysics for making face recognition al- gorithms more explainable. InProceedings of the European Conference on Computer Vision (ECCV), pages 252–270, 2018. 2 11

    2018

  51. [51]

    Eye movement-driven defense against iris print-attacks

    Ioannis Rigas and Oleg V Komogortsev. Eye movement-driven defense against iris print-attacks. Pattern Recognition Letters, 68:316–326, 2015. 3

    2015

  52. [52]

    New support vector algorithms.Neural computation, 12(5):1207–1245,

    Bernhard Sch ¨olkopf, Alex J Smola, Robert C Williamson, and Peter L Bartlett. New support vector algorithms.Neural computation, 12(5):1207–1245,

  53. [53]

    D-netpad: An explain- able and interpretable iris presentation attack detector

    Renu Sharma and Arun Ross. D-netpad: An explain- able and interpretable iris presentation attack detector. In2020 IEEE international joint conference on bio- metrics (IJCB), pages 1–10. IEEE, 2020. 3

    2020

  54. [54]

    Dinov3.arXiv preprint arXiv:2508.10104, 2025

    Oriane Sim ´eoni, Huy V V o, Maximilian Seitzer, Fed- erico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamonjisoa, et al. Dinov3.arXiv preprint arXiv:2508.10104, 2025. 3, 4

    Pith/arXiv arXiv 2025

  55. [55]

    Benchmarking foun- dation models for zero-shot biometric tasks.arXiv preprint arXiv:2505.24214, 2025

    Redwan Sony, Parisa Farmanifard, Hamzeh Alzwairy, Nitish Shukla, and Arun Ross. Benchmarking foun- dation models for zero-shot biometric tasks.arXiv preprint arXiv:2505.24214, 2025. 2, 4

    Pith/arXiv arXiv 2025

  56. [56]

    Tapia, L ´azaro Janier Gonz ´alez-Soler, and Christoph Busch

    Juan E. Tapia, L ´azaro Janier Gonz ´alez-Soler, and Christoph Busch. Towards iris presentation attack de- tection with foundation models. In2025 IEEE 19th International Conference on Automatic Face and Ges- ture Recognition (FG), pages 1–5, 2025. 2, 4

    2025

  57. [57]

    Saliency revisited: Analysis of mouse movements versus fixations

    Hamed R Tavakoli, Ahmed Pishe, Esa Rahtu, and Janne Heikkil¨a. Saliency revisited: Analysis of mouse movements versus fixations. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6354–6362, 2017. 3

    2017

  58. [58]

    Visualising spatio-temporal gaze characteristics for exploratory data analysis in clinical fetal ultrasound scans

    Clare Teng, Harshita Sharma, Lior Drukker, Aris T Papageorghiou, and J Alison Noble. Visualising spatio-temporal gaze characteristics for exploratory data analysis in clinical fetal ultrasound scans. In2022 Symposium on Eye Tracking Research and Applica- tions, pages 1–6, 2022. 4

    2022

  59. [59]

    Iris liveness detection competition (livdet- iris)–the 2023 edition

    Patrick Tinsley, Sandip Purnapatra, Mahsa Mitcheff, Aidan Boyd, Colton Crum, Kevin Bowyer, Patrick Flynn, Stephanie Schuckers, Adam Czajka, Meiling Fang, et al. Iris liveness detection competition (livdet- iris)–the 2023 edition. In2023 IEEE International Joint Conference on Biometrics (IJCB), pages 1–10. IEEE, 2023. 2

    2023

  60. [60]

    Understanding lin- ear probing then fine-tuning language models from ntk perspective.Advances in Neural Information Process- ing Systems, 37:139786–139822, 2024

    Akiyoshi Tomihari and Issei Sato. Understanding lin- ear probing then fine-tuning language models from ntk perspective.Advances in Neural Information Process- ing Systems, 37:139786–139822, 2024. 5

    2024

  61. [62]

    Assessment of iris recognition reliability for eyes affected by ocular pathologies

    Mateusz Trokielewicz, Adam Czajka, and Piotr Ma- ciejewicz. Assessment of iris recognition reliability for eyes affected by ocular pathologies. In2015 IEEE 7th International Conference on Biometrics Theory, Applications and Systems (BTAS), pages 1–6. IEEE,

  62. [63]

    Presentation attack detection for cadaver iris

    Mateusz Trokielewicz, Adam Czajka, and Piotr Ma- ciejewicz. Presentation attack detection for cadaver iris. In2018 IEEE 9th international conference on biometrics theory, applications and systems (BTAS), pages 1–10. IEEE, 2018. 2

    2018

  63. [64]

    Perception of image features in post- mortem iris recognition: Humans vs machines

    Mateusz Trokielewicz, Adam Czajka, and Piotr Ma- ciejewicz. Perception of image features in post- mortem iris recognition: Humans vs machines. In 2019 IEEE 10th International Conference on Biomet- rics Theory, Applications and Systems (BTAS), pages 1–8. IEEE, 2019. 2

    2019

  64. [65]

    Post-mortem iris recognition with deep- learning-based image segmentation.Image and Vision Computing, 94:103866, 2020

    Mateusz Trokielewicz, Adam Czajka, and Piotr Ma- ciejewicz. Post-mortem iris recognition with deep- learning-based image segmentation.Image and Vision Computing, 94:103866, 2020. 3

    2020

  65. [66]

    Siglip 2: Multilingual vision-language encoders with improved semantic un- derstanding, localization, and dense features.arXiv preprint arXiv:2502.14786, 2025

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision-language encoders with improved semantic un- derstanding, localization, and dense features.arXiv preprint arXiv:2502.14786, 2025. 3, 4

    Pith/arXiv arXiv 2025

  66. [67]

    Synthe- sis of large realistic iris databases using patch-based sampling

    Zhuoshi Wei, Tieniu Tan, and Zhenan Sun. Synthe- sis of large realistic iris databases using patch-based sampling. In2008 19th International Conference on Pattern Recognition, pages 1–4. IEEE, 2008. 3

    2008

  67. [68]

    David Yambay, Benedict Becker, Naman Kohli, Daksha Yadav, Adam Czajka, Kevin W. Bowyer, Stephanie Schuckers, Richa Singh, Mayank Vatsa, Afzel Noore, Diego Gragnaniello, Carlo Sansone, Luisa Verdoliva, Lingxiao He, Yiwei Ru, Haiqing Li, Nianfeng Liu, Zhenan Sun, and Tieniu Tan. LivDet- Iris 2017 – Iris Liveness Detection Competition 2017. In2017 IEEE Inter...

    2017

  68. [69]

    LivDet-Iris 2015 – Iris Liveness Detection Competition 2015

    David Yambay, Brian Walczak, Stephanie Schuckers, and Adam Czajka. LivDet-Iris 2015 – Iris Liveness Detection Competition 2015. In2017 IEEE Interna- tional Conference on Identity, Security and Behavior Analysis (ISBA), pages 1–6, 2017. 3

    2015

  69. [70]

    Human gaze assisted artificial intelligence: A review

    Ruohan Zhang, Akanksha Saran, Bo Liu, Yifeng Zhu, Sihang Guo, Scott Niekum, Dana Ballard, and Mary Hayhoe. Human gaze assisted artificial intelligence: A review. InIJCAI: Proceedings of the Conference, page 4951. NIH Public Access, 2020. 2

    2020

  70. [71]

    Cross-device iris pre- 12 sentation attack detection based on image-text multi- modal alignment

    Tian Zhang, Hui Zhang, Xiuying Wu, Hang Zou, Dexin Zhang, and Jing Liu. Cross-device iris pre- 12 sentation attack detection based on image-text multi- modal alignment. InProceedings of the 2024 2nd Asia Symposium on Image and Graphics, page 199–206, New York, NY , USA, 2025. Association for Comput- ing Machinery. 2

    2024

  71. [72]

    Learn- ing from models beyond fine-tuning.Nature Machine Intelligence, 7(1):6–17, 2025

    Hongling Zheng, Li Shen, Anke Tang, Yong Luo, Han Hu, Bo Du, Yonggang Wen, and Dacheng Tao. Learn- ing from models beyond fine-tuning.Nature Machine Intelligence, 7(1):6–17, 2025. 5 13

    2025