Pith. sign in

REVIEW 4 major objections 5 minor 132 references

Gaze3P: Gaze-Based Prediction of User-Perceived Privacy

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Eye gaze can reveal how private a person finds an image, and the prediction can be used to set differential privacy budgets.

desk verdict Gaze3P's new dataset is a genuine contribution, but the headline RQ3 claim—that gaze predictions can set DP budgets and beat prior work—is not actually tested; Table 4 runs on users' ground-truth ratings, not on gaze-model outputs. read the letter →

arxiv 2507.00596 v2 pith:3ZJ3NKWV submitted 2025-07-01 cs.HC cs.CR

classification cs.HCcs.CR
keywords gazeeyetrackinguser-perceivedprivacyquantificationdifferentialpersonalizedprivacy-preservingtechniques
verification ladder T0 review T1 audit T2 compute T3 formal

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 introduces Gaze3P, a dataset of eye-gaze recordings from 100 participants viewing 1,000 privacy-relevant images, each with a self-reported privacy rating. Using this dataset, the authors train machine-learning models that predict a viewer's perceived privacy from gaze alone, reporting accuracy well above chance. They then map the predicted privacy levels to the epsilon parameter of differential privacy, comparing four mapping functions and showing improved data utility over static and random baselines. If the central claim holds, gaze offers a continuous, implicit measure of subjective privacy that can be used to personalize privacy guarantees without explicit user input.

What carries the argument

The load-bearing identity is the explicit bijection between a privacy budget $\varepsilon$ and the maximal advantage of an adversary distinguishing two adjacent datasets, $\mathrm{adv}_\mathcal{A}(\varepsilon) = (e^\varepsilon - 1)/(e^\varepsilon + 1)$, which the paper inverts as $f(l) = \log((1+g(l))/(1-g(l)))$ to turn a 1-7 privacy rating $l$ into an $\varepsilon$ value. Around this identity, the paper wraps four candidate mapping functions—linear, exponential, sequential, and sigmoid—that encode different cognitive models of how people translate perceived sensitivity into a privacy-loss tolerance. The other machinery is the gaze-feature pipeline: raw eye-tracking streams are reduced to fixations, saccades, and pupil dilation, then passed to standard classifiers (decision trees, SVMs, logistic regression, random forests, k-nearest neighbours, and a transformer), so the prediction of perceived privacy is driven entirely by implicit gaze behaviour.

What would settle it

Re-run the identical eye-tracking protocol with participants who are not told the study concerns privacy and who do not rate images; if privacy classifiers trained on the primed Gaze3P data do not beat chance on the unprimed gaze, the claim that gaze is an implicit indicator of perceived privacy would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that human eye gaze is an implicit and dynamic indicator of user-perceived privacy. The paper supports this with Gaze3P, a dataset of gaze data from 100 participants viewing 1,000 images with private and safe attributes, and with models that predict binary privacy perception, seven-level privacy ratings, contextual ratings, and the specific private attribute being viewed, all from gaze features such as fixations, saccades, and pupil dilation. The paper further claims that gaze-based predictions can be mapped to differential privacy budgets through the adversary-advantage identity $\mathrm{adv}_\mathcal{A}(\varepsilon) = (e^\varepsilon - 1)/(e^\varepsilon + 1)$, and that the resulting $\varepsilon$ values, when substituted into existing personalized DP mechanisms, yield better utility than randomly chosen or uniformly worst-case budgets while aligning with user expectations.

Load-bearing premise

The experiments primed participants on privacy and asked them to imagine the images were their own, so the gaze patterns may reflect the explicit rating task rather than an implicit privacy judgment that happens naturally outside the lab.

Editorial extensions

If this is right

  • Gaze-based predictions can replace questionnaires as an implicit, dynamic source of privacy perception, updating continuously as the viewed content changes.
  • Choosing $\varepsilon$ through the gaze-derived mappings improves utility of DP-protected analytics and learning tasks compared with static worst-case or random budgets, on the benchmarks considered.
  • The 30 Hz downsampling result implies commodity webcams may be sufficient to deploy gaze-based privacy prediction in practice.
  • Demographic context and privacy expertise shape gaze behaviour and add predictive signal, so personalized models benefit from user metadata.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next experiment is a between-subject design with unprimed participants to separate intrinsic gaze indicators from task-induced attention; the paper itself notes this limitation.
  • The adversary-advantage mapping could be reused to convert any behavioural signal—not just gaze—into $\varepsilon$ values, provided the signal can be calibrated to a privacy rating scale.
  • The same pipeline could set other hand-picked privacy parameters, such as gradient clipping norms in federated learning or $k$ in $k$-anonymity, as the paper lists but does not implement.
  • Because residual gaze maps after subtracting saliency are far less predictive, gaze-based privacy prediction may work best on images where private content coincides with visually salient regions; testing on more diverse image compositions would clarify this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces Gaze3P, a dataset of gaze recordings from 100 participants viewing 1,000 privacy-relevant images with self-reported privacy ratings, and uses it to train machine learning models for several tasks: binary privacy perception, privacy level perception, contextual privacy perception, private attribute recognition, privacy expertise prediction, and privacy-aware gaze identification. The authors report modest gaze-only accuracy (e.g., person-independent binary accuracy 0.64, level accuracy 0.34) and much higher accuracy when stimulus features are added. They then propose four mapping functions from perceived privacy levels to differential privacy epsilon values and evaluate these on PDP benchmarks for data analysis and learning, reporting better utility than a random baseline and static worst-case DP. A follow-up user study asks participants whether reconstructed images match their expectations. The paper concludes that gaze is an implicit and dynamic indicator of user-perceived privacy and that gaze-based predictions can personalize DP epsilon budgets while outperforming previous work.

Significance. If the central claims held, Gaze3P would be a valuable community resource: it is the first large-scale gaze dataset for perceived privacy, covers a diverse stimulus set, and proposes a concrete route from behavioural signals to personalized differential privacy parameters. The paper's strengths include the breadth of tasks, the use of standard baselines, and the explicit comparison with the VISPR and PrivacEye benchmarks. However, the current evidence does not support the strongest claims: the gaze-only accuracies are close to chance for several tasks, the DP evaluation never feeds gaze-model predictions into the epsilon mapping, and the 'implicit' interpretation is weakened by the experiment's explicit privacy framing, which the authors themselves acknowledge in Section 7.1. The dataset and task taxonomy are still likely to be useful to the community, but the paper needs substantial revision to align its claims with its evidence.

major comments (4)
  1. [Section 6.2, Table 4] RQ3 is not tested end-to-end. The privacy budgets in the PDP evaluation are obtained by applying the mapping functions of Section 6.1 to the users' ground-truth privacy ratings l, not to the outputs of the gaze-based classifiers trained in Section 5. The text says 'user-specified privacy preferences inferred from gaze,' but no experiment feeds a gaze-predicted level through f(l): the 'Gaze3P' columns in Table 4 are therefore a self-reported-rating condition. The claim that 'our gaze-based approach outperforms previous work' (Section 1) is unsupported by this table, because the improvement over random could be due to the per-user ground-truth information rather than to the gaze signal. Please add a condition that uses held-out gaze predictions from Section 5 (person-specific and person-independent) and report variance and paired significance tests. The follow-up user study likewise presents participants with their previously selected privacy levels, not model predictions.
  2. [Section 1, Table 1] The abstract and Section 1 claim models trained 'solely from gaze' achieve 'high accuracy,' but the gaze-only person-independent results are modest: the best binary accuracy is 0.64 (chance 0.5) and the best level accuracy is 0.34 (chance 0.14). The high accuracies 0.97/0.87 in Table 1 come from the 'Gaze + Stimuli' rows, which use image features as input and therefore do not support the gaze-only claim. Please either rephrase the headline claims for RQ2 to report gaze-only performance, or provide additional analysis showing that the gaze+stimuli results are not driven by the stimulus features alone (the PrivacEye comparison in Table 1 provides only a partial baseline).
  3. [Section 7.1, Section 4] The experiment design itself undermines the 'implicit' claim in RQ1. Participants were told the study was about privacy, were instructed to imagine the images as their own phone gallery, and were asked to rate each image's privacy (Section 4). As the Limitations section admits, this priming 'likely influenced cognitive processing and visual attention,' so the gaze patterns may reflect the explicit rating task rather than an implicit, naturally occurring privacy judgment. The abstract and conclusion nevertheless state that gaze is an 'implicit and dynamic indicator.' Please temper the claims to privacy-aware scenarios or add a between-subject neutral condition to support the stronger interpretation.
  4. [Section 6.1, Table 4] The PDP utility comparison lacks statistical grounding. Table 4 reports single numbers with no variance, confidence intervals, or significance tests, and the 'random' baseline is not specified (distribution over epsilon values, whether per-user or global). The mapping functions contain free parameters k, epsilon_min, and epsilon_max, and no sensitivity analysis is provided for these choices. Consequently, the conclusion that the proposed mappings 'yield better utility than the random benchmark' (Section 6.2) is not yet supported. Please report repeated-run statistics and, at minimum, a sensitivity analysis over k and epsilon bounds.
minor comments (5)
  1. [Section 1] There is a typo 'e,g,' for 'e.g.' and duplicated reference keys in the phrase '[21, 21, 28, 28]'; the latter suggests a citation cleanup is needed.
  2. [Section 5.1, Table 1] In the down-sampling block, the DT binary accuracy is 0.48, below the 0.5 chance level, while the text says results exceed chance levels; please clarify this exception.
  3. [Section 5.2, Table 3] For privacy expertise prediction, SVM and LR achieve 0.45 and 0.46, at or below chance, but the text says 'results exceed the chance levels' without discussing these model-specific failures.
  4. [Section 5.2] The sentence 'K-means clustering yields the lowest silhouette score with 12 distinct gaze behaviour profiles' is ambiguous, since a lower silhouette score indicates weaker clustering; please clarify the intended comparison or metric.
  5. [Section 6.2] In the user expectation alignment study, the statement that 'the subsample shown to participants was representative' is not supported with demographic or rating comparisons, and the details of the MLDoctor reconstruction pipeline are missing; please provide these.

Circularity Check

1 steps flagged · score 6.0 of 10

RQ3's 'gaze-based' DP evaluation bypasses the gaze model: Table 4 sets epsilon from users' ground-truth privacy ratings while labeling them 'inferred from gaze.'

  1. fitted input called prediction [Section 6.2, 'Empirical Evaluation' and 'User expectation alignment', Table 4]
    "The only alteration involves replacing the randomly generated privacy-level inputs (ε) commonly used in prior work with user-specified privacy preferences inferred from gaze. ... For the validation study, each participant was shown the same set of images they had rated previously during the initial dataset collection (the search task), along with their user-specified privacy levels."

    The DP utility experiment computes ε by applying the Section 6.1 mapping functions to the users' own privacy ratings l, which are the exact ground-truth labels the Section 5 gaze models were trained to predict. No experimental step feeds a gaze-model output through f(l); the follow-up validation likewise presents the participants' previously given ratings, not model predictions.

full rationale

The manuscript's RQ1/RQ2 core is a genuine supervised-learning study: gaze features are extracted from a purpose-built 100-participant dataset and held-out predictions of privacy ratings are evaluated against those ratings; this is a standard, non-circular empirical pipeline. The Limitations section even concedes that the privacy-priming context shapes gaze, which is a generalization threat, not circularity. The circularity concerns are concentrated in RQ3. Section 6.2 claims to replace random ε with 'user-specified privacy preferences inferred from gaze,' but the actual evaluation (Table 4 and the follow-up study) computes ε from the users' own previously collected privacy ratings—i.e., the exact labels the gaze models were trained to predict. No gaze-model output is fed through the mapping f(l), and the user study presents participants' previously selected levels, not predictions. Thus the headline claim 'our gaze-based approach outperforms previous work' reduces, by construction, to 'per-user ground-truth ratings outperform random ε,' which is an artifact of using the label as the input. This is a partial circularity: it does not invalidate the dataset or the supervised gaze-classification results, but it does mean the paper's only demonstrated DP-utility improvement does not involve gaze predictions at all.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities. The free parameters are the hand-chosen sigmoid steepness and epsilon bounds; the axioms are the operationalization of perceived privacy via Likert ratings, the generalizability of primed gaze behavior, the standard adversary-advantage identity, the cognitive plausibility of the four mappings, and the adequacy of the chosen utility benchmarks.

free parameters (2)
  • Sigmoid steepness k = 1.5
    Chosen by hand in Section 6.1.4 to represent 'moderate steepness'; no sensitivity analysis is reported, so the utility results in Table 4 may depend on this choice.
  • DP epsilon bounds (epsilon_min, epsilon_max) = 0.1, 5
    Set in Section 6.1 and Fig 7 as application-specific bounds, following [94]; they are not fitted to the data but affect the mapping scale and the utility comparison.
assumptions (5)
  • domain assumption Self-reported privacy ratings on a 1-7 Likert scale are a valid operationalization of user-perceived privacy.
    The dataset ground truth and all ML labels are these ratings (Section 4, Experiment design); the paper itself notes the limitations of questionnaires in Section 1, creating tension between the criticism and the ground truth.
  • domain assumption Gaze behavior collected in a privacy-primed, rating-eliciting laboratory setting generalizes to implicit privacy perception in natural interactions.
    The experiment primes participants on privacy and asks for explicit ratings; Section 7.1 acknowledges this contextual bias restricts generalizability to privacy-aware scenarios.
  • standard math The adversary advantage advA(epsilon) = (e^epsilon - 1)/(e^epsilon + 1) is the appropriate bridge between the DP budget and user-perceived privacy loss.
    Derived from the epsilon-DP definition in Section 2 and Appendix E; the mapping functions in Section 6.1 are built on this identity.
  • ad hoc to paper The four proposed mapping functions (linear, exponential, sequential, sigmoid) capture the cognitive relationships between privacy ratings and acceptable privacy loss.
    These are design choices motivated by cognitive theories (Appendix D), but they are not derived from or validated against independent behavioral data; the paper recommends selecting among them post hoc per attribute (Section 6.2).
  • domain assumption The chosen benchmarks (count, median, min queries; linear regression, weighting, sampling) measure the utility of the obfuscated data adequately.
    Adopted from Jorgensen et al. [61] and Boenisch et al. [10, 11]; utility is measured only through these tasks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Gaze3P: Gaze-Based Prediction of User-Perceived Privacy." pith.science (2026). https://pith.science/paper/3ZJ3NKWV

@misc{pith2026250700596,
  author       = {Pith},
  title        = {Pith review of: Gaze3P: Gaze-Based Prediction of User-Perceived Privacy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3ZJ3NKWV}},
  note         = {Machine review of arXiv:2507.00596}
}
read the original abstract

Privacy is a highly subjective concept and perceived variably by different individuals. Previous research on quantifying user-perceived privacy has primarily relied on questionnaires. Furthermore, applying user-perceived privacy to optimise the parameters of privacy-preserving techniques (PPT) remains insufficiently explored. To address these limitations, we introduce Gaze3P -- the first dataset specifically designed to facilitate systematic investigations into user-perceived privacy. Our dataset comprises gaze data from 100 participants and 1,000 stimuli, encompassing a range of private and safe attributes. With Gaze3P, we train a machine learning model to implicitly and dynamically predict perceived privacy from human eye gaze. Through comprehensive experiments, we show that the resulting models achieve high accuracy. Finally, we illustrate how predicted privacy can be used to optimise the parameters of differentially private mechanisms, thereby enhancing their alignment with user expectations.

Figures

Figures reproduced from arXiv: 2507.00596 by the authors.

Figure 1
Figure 1. Users deal with various types of information in daily life that can have vastly different privacy requirements, e.g., [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Sample images from the VISPR dataset with safe (e.g. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative example of the inter- and intra-person [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Qualitative example of learning effect: Examples [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative example for visual privacy observations: [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: The left figure shows the distribution of the perceived privacy levels [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Random samples of the reconstructed stimuli (first [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Demographics’ distribution of the number of male [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: The eye-tracker setup 22 A.3 Experiment Design 23 A sample trial is conducted as shown in [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: A sample trial flow-chart 24 The stimuli were gathered from the validation and test sets of 25 the VISPR dataset [101]. Every block contained writing-based and 26 human stimuli to avoid bias, with a distribution of private attributes 27 as shown in [PITH_FULL_IMAGE:f…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

132 extracted references · 57 canonical work pages

  1. [1]

    Yaseem Abdrabou, Mariam Hassib, Shuqin Hu, Ken Pfeuffer, Mohamed Khamis, Andreas Bulling, and Florian Alt. 2024. EyeSeeIdentity: Exploring Natural Gaze Behavior for Implicit User Identification during Photo Viewing. (2024)

  2. [2]

    Krishna Acharya, Franziska Boenisch, Rakshit Naidu, and Juba Ziani. 2024. Personalized differential privacy for ridge regression. arXiv preprint arXiv:2401.17127 (2024)

  3. [3]

    Zahra Aivazpour, Rohit Valecha, and Raghav H Rao. 2017. Unpacking privacy paradox: a dual process theory approach. (2017)

  4. [4]

    Mohammad Alaggan, Sébastien Gambs, and Anne-Marie Kermarrec. 2015. Het- erogeneous differential privacy. arXiv preprint arXiv:1504.06998 (2015)

  5. [5]

    Bonnie Brinton Anderson, Jeffrey L Jenkins, Anthony Vance, C Brock Kirwan, and David Eargle. 2016. Your memory is working against you: How eye tracking and memory explain habituation to security warnings. Decision Support Systems 92 (2016), 3–13

  6. [6]

    DEMOKRITOS

    Ion Androutsopoulos, Georgios Paliouras, and Eirinaios Michelakis. 2004.Learn- ing to filter unsolicited commercial e-mail . Vol. 2004. " DEMOKRITOS", National Center for Scientific Research

  7. [7]

    Apple. 2025. Privacy Labels. https://www.apple.com/privacy/labels/ Accessed: 2025-02-18

  8. [8]

    Anonymous Author(s). 2025. Our Implementation Data to Gaze3P. To be published upon acceptance

Show all 132 references
  1. [9]

    Ardion Beldad, Menno De Jong, and Michaël Steehouder. 2011. I trust not therefore it must be risky: Determinants of the perceived risks of disclosing personal data for e-government transactions. Computers in Human Behavior 27, 6 (2011), 2233–2242

  2. [10]

    Franziska Boenisch, Christopher Mühl, Adam Dziedzic, Roy Rinberg, and Nico- las Papernot. 2023. Have it your way: Individualized Privacy Assignment for DP-SGD. In Advances in Neural Information Processing Systems , A. Oh, T. Nau- mann, A. Globerson, K. Saenko, M. Hardt, and S....

  3. [11]

    Franziska Boenisch, Christopher Mühl, Roy Rinberg, Jannis Ihrig, and Adam Dziedzic. 2022. Individualized PATE: Differentially private machine learning with individual privacy guarantees. arXiv preprint arXiv:2202.10517 (2022)

  4. [12]

    Efe Bozkir, Onur Günlü, Wolfgang Fuhl, Rafael F Schaefer, and Enkelejda Kasneci

  5. [13]

    Efe Bozkir, Süleyman Özdel, Mengdi Wang, Brendan David-John, Hong Gao, Kevin Butler, Eakta Jain, and Enkelejda Kasneci. 2023. Eye-tracked Virtual Reality: A Comprehensive Survey on Methods and Privacy Challenges

  6. [14]

    Stephanie Brams, Gal Ziv, Oron Levin, Jochim Spitz, Johan Wagemans, A Mark Williams, and Werner F Helsen. 2019. The relationship between gaze behavior, expertise, and performance: A systematic review. Psychological bulletin 145, 10 (2019), 980

  7. [15]

    Andreas Bulling and Daniel Roggen. 2011. Recognition of Visual Memory Recall Processes Using Eye Movement Analysis. In ACM UbiComp. 455–464. https://doi.org/10.1145/2030112.2030172

  8. [16]

    Paul-Christian Bürkner and Matti Vuorre. 2019. Ordinal regression models in psychology: A tutorial. Advances in Methods and Practices in Psychological Science 2, 1 (2019), 77–101

  9. [17]

    Virginio Cantoni, Chiara Galdi, Michele Nappi, Marco Porta, and Daniel Ric- cio. 2015. GANT: Gaze analysis technique for human identification. Pattern Recognition 48, 4 (2015), 1027–1038

  10. [18]

    Benjamin T Carter and Steven G Luke. 2020. Best practices in eye tracking research. International Journal of Psychophysiology 155 (2020), 49–62

  11. [19]

    Yimin Chen, Tao Li, Rui Zhang, Yanchao Zhang, and Terri Hedgpeth. 2018. EyeTell: Video-Assisted Touchscreen Keystroke Inference from Eye Movements. In 2018 IEEE Symposium on Security and Privacy (SP) . 144–160. https://doi.org/ 10.1109/SP.2018.00010

  12. [20]

    Bryan H Choi. 2014. A Prospect Theory of Privacy. Idaho L. Rev. 51 (2014), 623

  13. [21]

    Mary J Culnan and Pamela K Armstrong. 1999. Information privacy concerns, procedural fairness, and impersonal trust: An empirical investigation. Organi- zation science 10, 1 (1999), 104–115

  14. [22]

    Rachel Cummings, Damien Desfontaines, David Evans, Roxana Geambasu, Yangsibo Huang, Matthew Jagielski, Peter Kairouz, Gautam Kamath, Sewoong Oh, Olga Ohrimenko, et al. 2023. Advancing differential privacy: Where we are now and future directions for real-world deployment. arXiv...

  15. [23]

    Rachel Cummings and David Durfee. 2020. Individual sensitivity preprocessing for data privacy. In Proceedings of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms. SIAM, 528–547

  16. [24]

    Rachel Cummings, Katrina Ligett, Aaron Roth, Zhiwei Steven Wu, and Juba Ziani. 2015. Accuracy for sale: Aggregating data with a variance constraint. In Proceedings of the 2015 conference on innovations in theoretical computer science . 317–324

  17. [25]

    Brendan David-John, Kevin Butler, and Eakta Jain. 2022. For your eyes only: Privacy-preserving eye-tracking datasets. In ACM ETRA. 1–6

  18. [26]

    Brendan David-John, Diane Hosfelt, Kevin Butler, and Eakta Jain. 2021. A privacy-preserving approach to streaming eye-tracking data. IEEE TVCG 27, 5 (2021), 2555–2565

  19. [27]

    Brendan David-John, Candace Peacock, Ting Zhang, T Scott Murdison, Hrvoje Benko, and Tanya R Jonker. 2021. Towards gaze-based prediction of the intent to interact in virtual reality. In ACM ETRA. 1–7

  20. [28]

    Tamara Dinev and Paul Hart. 2006. An extended privacy calculus model for e-commerce transactions. Information systems research 17, 1 (2006), 61–80

  21. [29]

    D Dittrich and E Kenneally. 2012. The Menlo Report: Ethical Principles Guiding Information and Communication Technology Research . Technical Report. U.S. Department of Homeland Security. https://doi.org/paper/2012_menlo_report_ actual_formatted

  22. [30]

    Essohanam Djeki, Jules Dégila, and Muhtar Hanif Alhassan. 2024. Reimagining Authentication: A User-Centric Two-Factor Authentication with Personalized Image Verification. In 2024 ASU International Conference in Emerging Technolo- gies for Sustainability and Intelligent Systems...

  23. [31]

    Cynthia Dwork. 2006. Differential Privacy. In Automata, Languages and Pro- gramming, Michele Bugliesi, Bart Preneel, Vladimiro Sassone, and Ingo Wegener (Eds.). Springer Berlin Heidelberg

  24. [32]

    Cynthia Dwork, Aaron Roth, et al. 2014. The algorithmic foundations of differ- ential privacy. Foundations and Trends® in Theoretical Computer Science 9, 3–4 (2014), 211–407

  25. [33]

    Alessandro D’Amelio, Sabrina Patania, Sathya Bursic, Vittorio Cuculo, and Giuseppe Boccignone. 2023. Using gaze for behavioural biometrics. Sensors 23, 3 (2023), 1262

  26. [34]

    Hamid Ebadi, David Sands, and Gerardo Schneider. 2015. Differential privacy: Now it’s getting personal. Acm Sigplan Notices 50, 1 (2015), 69–81

  27. [35]

    Maria K Eckstein, Belén Guerra-Carrillo, Alison T Miller Singley, and Silvia A Bunge. 2017. Beyond eye gaze: What else can eyetracking reveal about cognition and cognitive development? Developmental cognitive neuroscience 25 (2017), 69–91

  28. [36]

    Yusra Elbitar, Michael Schilling, Trung Tin Nguyen, Michael Backes, and Sven Bugiel. 2021. Explanation beats context: The effect of timing & rationales on users’ runtime permission decisions. In 30th USENIX Security Symposium (USENIX Security 21). 785–802

  29. [37]

    Mayar Elfares, Zhiming Hu, Pascal Reisert, Andreas Bulling, and Ralf Küsters

  30. [38]

    Mayar Elfares, Pascal Reisert, Zhiming Hu, Ralf Küsters, and Andreas Bulling

  31. [39]

    Mayar Elfares, Pascal Reisert, Ralf Küsters, and Andreas Bulling. 2025. QualitEye: Public and Privacy-preserving Gaze Data Quality Verification. arXiv preprint arXiv:2506.05908 (2025)

  32. [40]

    Myrthe Faber, Robert Bixler, and Sidney K D’Mello. 2018. An automated be- havioral measure of mind wandering during computerized reading. Behavior Research Methods 50, 1 (2018), 134–150

  33. [41]

    Alireza Fallah, Ali Makhdoumi, Asuman Ozdaglar, et al. 2022. Bridging central and local differential privacy in data acquisition mechanisms. Advances in Neural Information Processing Systems 35 (2022), 21628–21639

  34. [42]

    Filippo Galli, Sayan Biswas, Kangsoo Jung, Tommaso Cucinotta, and Catuscia Palamidessi. 2022. Group privacy for personalized federated learning. https: //doi.org/10.48550/ARXIV.2206.03396

  35. [43]

    Mohamad Gharib. 2024. Towards a Heuristic Model for Usable Privacy.. In RCIS Workshops

  36. [44]

    Arpita Ghosh and Aaron Roth. 2011. Selling privacy at auction. In Proceedings of the 12th ACM conference on Electronic commerce . 199–208

  37. [45]

    Google. 2021. Federated Learning of Cohorts (FLoC) Whitepaper. https://raw.githubusercontent.com/google/ads-privacy/master/proposals/ FLoC/FLOC-Whitepaper-Google.pdf Accessed: 2025-02-18

  38. [46]

    Google. 2025. Google Safe Browsing. https://safebrowsing.google.com/ Ac- cessed: 2025-02-18

  39. [47]

    Céline Gressel, Rebekah Overdorf, Inken Hagenstedt, Murat Karaboga, Helmut Lurtz, Michael Raschke, and Andreas Bulling. 2023. Privacy-Aware Eye Tracking: Challenges and Future Directions.IEEE Pervasive Computing 22, 1 (2023), 95–102. https://doi.org/10.1109/MPRV.2022.3228660

  40. [48]

    Michele Guerra, Roberto Milanese, Madalina G Ciobanu, Michele Deodato, and Fausto Fasano. 2023. Seeing is Believing: Assessing and Enhancing Android Privacy Indicators Through Eye-Tracking Analysis. In International Conference on Information Systems Security and Privacy . Spri...

  41. [49]

    Michele Guerra, Roberto Milanese, Michele Deodato, Vittorio Perozzi, Fausto Fasano, et al. 2024. Visual Attention and Privacy Indicators in Android: Insights from Eye Tracking.. In ICISSP. 320–329

  42. [50]

    Michele Guerra, Simone Scalabrino, Fausto Fasano, and Rocco Oliveto. 2023. An empirical study on the effectiveness of privacy indicators. IEEE Transactions on Software Engineering 49, 10 (2023), 4610–4623. 14 Gaze3P: Gaze-Based Prediction of User-Perceived Privacy Proceedings ...

  43. [51]

    Eddie Harmon-Jones and Cindy Harmon-Jones. 2012. Cognitive dissonance theory. Handbook of motivation science 71 (2012)

  44. [52]

    Glenn W Harrison and J Todd Swarthout. 2019. Eye-tracking and economic theories of choice under risk. Journal of the Economic Science Association 5, 1 (2019), 26–37

  45. [53]

    Roy S Hessels, Antje Nuthmann, Marcus Nyström, Richard Andersson, Dieder- ick C Niehorster, and Ignace TC Hooge. 2025. The fundamentals of eye tracking part 1: The link between theory and research question. Behavior Research Methods 57, 1 (2025), 1–18

  46. [54]

    Ignace TC Hooge, Antje Nuthmann, Marcus Nyström, Diederick C Niehorster, Gijs A Holleman, Richard Andersson, and Roy S Hessels. 2025. The fundamentals of eye tracking part 2: From research question to operationalization. Behavior Research Methods 57, 2 (2025), 73

  47. [55]

    Roberto Hoyle, Luke Stark, Qatrunnada Ismail, David Crandall, Apu Kapadia, and Denise Anthony. 2020. Privacy norms and preferences for photos posted online. ACM Transactions on Computer-Human Interaction (TOCHI) 27, 4 (2020), 1–27

  48. [56]

    Justin Hsu, Marco Gaboardi, Andreas Haeberlen, Sanjeev Khanna, Arjun Narayan, Benjamin C Pierce, and Aaron Roth. 2014. Differential privacy: An economic method for choosing epsilon. In 2014 IEEE 27th Computer Security Foundations Symposium. IEEE, 398–410

  49. [57]

    Vincent C Hu, D Richard Kuhn, David F Ferraiolo, and Jeffrey Voas. 2015. Attribute-based access control. Computer 48, 2 (2015), 85–88

  50. [58]

    Michael Xuelin Huang, Jiajia Li, Grace Ngai, and Hong Va Leong. 2016. Stress- click: Sensing stress from gaze-click patterns. In Proceedings of the 24th ACM international conference on Multimedia . 1395–1404

  51. [59]

    Athina Ioannou, Iis Tussyadiah, Graham Miller, Shujun Li, and Mario Weick

  52. [60]

    Carl Anderson Johnson. 1974. Privacy as personal control. Man-environment interactions: evaluations and applications: part 2 (1974), 83–100

  53. [61]

    Zach Jorgensen, Ting Yu, and Graham Cormode. 2015. Conservative or liberal? Personalized differential privacy. In 2015 IEEE 31St international conference on data engineering. IEEE, 1023–1034

  54. [62]

    Daniel Kahneman and Amos Tversky. 2013. Prospect theory: An analysis of decision under risk. In Handbook of the fundamentals of financial decision making: Part I. World Scientific, 99–127

  55. [63]

    PloS one 16, 8 (2021), e0256822

    Privacy nudges for disclosure of personal information: A systematic literature review and meta-analysis. PloS one 16, 8 (2021), e0256822

  56. [64]

    Raptis, Mohamed Khamis, and Florian Alt

    Christina Katsini, Yasmeen Abdrabou, George E. Raptis, Mohamed Khamis, and Florian Alt. 2020. The Role of Eye Gaze in Security and Privacy Applications: Survey and Future HCI Research Directions. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (...

  57. [65]

    Murat Kezer, Tobias Dienlin, and Lemi Baruh. 2022. Getting the privacy calculus right: Analyzing the relations between privacy concerns, expected benefits, and self-disclosure using response surface analysis. Cyberpsychology: Journal of Psychosocial Research on Cyberspace 16, 4 (2022)

  58. [66]

    Wonjae Kim, Bokyung Son, and Ildoo Kim. 2021. Vilt: Vision-and-language trans- former without convolution or region supervision. In International conference on machine learning. PMLR, 5583–5594

  59. [67]

    Peter Kairouz, H Brendan McMahan, Brendan Avent, Bellet, et al. 2021. Advances and open problems in federated learning. Foundations and Trends® in Machine Learning 14, 1–2 (2021), 1–210

  60. [68]

    Leanne K Knobloch. 2008. Uncertainty reduction theory. Engaging theories in interpersonal communication (2008), 133–144

  61. [69]

    Nitin Kohli and Paul Laskowski. 2018. Epsilon voting: Mechanism design for parameter selection in differential privacy. In 2018 IEEE Symposium on Privacy- A ware Computing (PAC). IEEE, 19–30

  62. [70]

    Spyros Kokolakis. 2017. Privacy attitudes and privacy behaviour: A review of current research on the privacy paradox phenomenon. Computers & security 64 (2017), 122–134

  63. [71]

    Bart P Knijnenburg, Reza Ghaiumy Anaraky, Daricia Wilkinson, Moses Namara, Yangyang He, David Cherry, and Erin Ash. 2022. User-Tailored Privacy

  64. [72]

    Matthias Kümmerer, Matthias Bethge, and Thomas S. A. Wallis. 2022. DeepGaze III: Modeling free-viewing human scanpaths with deep learn- ing. Journal of Vision 22, 5 (04 2022), 7–7. https://doi.org/10.1167/jov.22.5. 7 arXiv:https://arvojournals.org/arvo/content_public/journal/j...

  65. [73]

    Benjamin Law, M Stella Atkins, Arthur E Kirkpatrick, and Alan J Lomax. 2004. Eye gaze patterns differentiate novice and experts in a virtual laparoscopic surgery training environment. In Proceedings of the 2004 symposium on Eye tracking research & applications . 41–48

  66. [74]

    Jaewoo Lee and Chris Clifton. 2011. How much is enough? choosing 𝜀 for differential privacy. In Information Security: 14th International Conference, ISC 2011, Xi’an, China, October 26-29, 2011. Proceedings 14 . Springer, 325–340

  67. [75]

    Kyle Krafka, Aditya Khosla, Petr Kellnhofer, Harini Kannan, Suchendra Bhan- darkar, Wojciech Matusik, and Antonio Torralba. 2016. Eye tracking for every- one. In IEEE ICPR. 2176–2184

  68. [76]

    Meng Li, Liehuang Zhu, Zijian Zhang, and Rixin Xu. 2017. Achieving differential privacy of trajectory data publishing in participatory sensing. Information Sciences 400 (2017), 1–13

  69. [77]

    Ninghui Li, Tiancheng Li, and Suresh Venkatasubramanian. 2006. t-closeness: Privacy beyond k-anonymity and l-diversity. In 2007 IEEE 23rd international conference on data engineering . IEEE, 106–115

  70. [78]

    Jonathan Liebers and Stefan Schneegass. 2020. Gaze-based authentication in virtual reality. In ACM Symposium on Eye Tracking Research and Applications . 1–2

  71. [79]

    Jingjie Li, Amrita Roy Chowdhury, Kassem Fawaz, and Younghyun Kim. 2021. {Kal𝜀ido}:{Real-Time} Privacy Control for{Eye-Tracking} Systems. In 30th USENIX Security Symposium. 1793–1810

  72. [80]

    Fan Liu, Zhiyong Cheng, Huilin Chen, Yinwei Wei, Liqiang Nie, and Mohan Kankanhalli. 2022. Privacy-preserving synthetic data generation for recommen- dation systems. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrie...

  73. [81]

    Yan Liu, Pei-Yun Hsueh, Jennifer Lai, Mirweis Sangin, Marc-Antoine Nussli, and Pierre Dillenbourg. 2009. Who is the expert? Analyzing gaze data to predict expertise level in collaborative applications. In2009 IEEE international conference on Multimedia and Expo . IEEE, 898–901

  74. [82]

    Yugeng Liu, Rui Wen, xinlei.he, Ahmed Salem, Zhikun Zhang, Michael Backes, Emiliano De Cristofaro, Mario Fritz, and Yang Zhang. 2022. ML-Doctor: Holistic Risk Assessment of Inference Attacks Against Machine Learning Models. (8 2022). https://doi.org/10.60882/cispa.24614037.v1

  75. [83]

    Ao Liu, Lirong Xia, Andrew Duchowski, Reynold Bailey, Kenneth Holmqvist, and Eakta Jain. 2019. Differential privacy for eye-tracking data. In Proceedings of the 11th ACM Symposium on Eye Tracking Research & Applications . 1–10

  76. [84]

    Shiqing Luo, Anh Nguyen, Chen Song, Feng Lin, Wenyao Xu, and Zhisheng Yan. 2020. OcuLock: Exploring Human Visual System for Authentication in Virtual Reality Head-mounted Display. In 2020 Network and Distributed System Security Symposium (NDSS). https://doi.org/10.14722/ndss.2...

  77. [85]

    Ashwin Machanavajjhala, Daniel Kifer, Johannes Gehrke, and Muthuramakr- ishnan Venkitasubramaniam. 2007. l-diversity: Privacy beyond k-anonymity. Acm transactions on knowledge discovery from data (tkdd) 1, 1 (2007), 3–es

  78. [86]

    Kelly D Martin and Patrick E Murphy. 2017. The role of data privacy in marketing. Journal of the Academy of Marketing Science 45 (2017), 135–155

  79. [87]

    Mingjie Lu and Zhenhua Liu. 2023. Improving Accuracy of Interactive Queries in Personalized Differential Privacy. In International Conference on Frontiers in Cyber Security. Springer, 141–159

  80. [88]

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. 2017. Communication-efficient learning of deep net- works from decentralized data. In Artificial intelligence and statistics . PMLR, 1273–1282

  81. [89]

    Yannic Meier and Nicole C Krämer. 2024. The privacy calculus revisited: an empirical investigation of online privacy decisions on between-and within- person levels. Communication Research 51, 2 (2024), 178–202

  82. [90]

    Meta. 2024. Providing Context on Sensitive or Misleading Con- tent. https://transparency.meta.com/en-gb/enforcement/taking-action/context- on-sensitive-misleading-content/ Accessed: 2025-02-18

  83. [91]

    Philipp K Masur. 2018. Situational privacy and self-disclosure: Communication processes in online environments . Springer

  84. [92]

    Milica Milosavljevic Mormann and Cary Frydman. 2016. The role of salience and attention in choice under risk: An experimental investigation. SSRN Electronic Journal 10 (2016)

  85. [93]

    Gilbert Mushure. 2017. Measuring Perception: The limitations of the question- naire methodology for gathering research data. (09 2017). https://doi.org/10. 13140/RG.2.2.24689.17766

  86. [94]

    Milad Nasr, Shuang Songi, Abhradeep Thakurta, Nicolas Papernot, and Nicholas Carlin. 2021. Adversary instantiation: Lower bounds for differentially private machine learning. In 2021 IEEE Symposium on security and privacy (SP) . IEEE, 866–882

  87. [95]

    Sounak Mondal, Zhibo Yang, Seoyoung Ahn, Dimitris Samaras, Gregory Zelin- sky, and Minh Hoai. 2023. Gazeformer: Scalable, effective and fast prediction of goal-directed human attention. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 1441–1450

  88. [96]

    Helen Nissenbaum. 2004. Privacy as contextual integrity.Wash. L. Rev.79 (2004), 119

  89. [97]

    Helen Nissenbaum. 2009. Privacy in context: Technology, policy, and the in- tegrity of social life. In Privacy in Context. Stanford University Press

  90. [98]

    Ben Niu, Yahong Chen, Boyang Wang, Zhibo Wang, Fenghua Li, and Jin Cao

  91. [99]

    Diederick C Niehorster, Marcus Nyström, Roy S Hessels, Richard Andersson, Jeroen S Benjamins, Dan Witzner Hansen, and Ignace TC Hooge. 2025. The fundamentals of eye tracking part 4: Tools for conducting an eye tracking study. Behavior Research Methods 57, 1 (2025), 46

  92. [100]

    Marcus Nyström, Ignace TC Hooge, Roy S Hessels, Richard Andersson, Dan Witzner Hansen, Roger Johansson, and Diederick C Niehorster. 2025. The fundamentals of eye tracking part 3: How to choose an eye tracker. Behavior Research Methods 57, 2 (2025), 67

  93. [101]

    Tribhuvanesh Orekondy, Bernt Schiele, and Mario Fritz. 2017. Towards a vi- sual privacy advisor: Understanding and predicting privacy risks in images. In Proceedings of the IEEE international conference on computer vision . 3686–3695

  94. [102]

    Sina Ostendorf, Silke M Müller, and Matthias Brand. 2020. Neglecting long-term risks: self-disclosure on social media and its relation to individual decision- making tendencies and problematic social-networks-use. Frontiers in Psychology 11 (2020), 543388

  95. [103]

    In IEEE INFOCOM 2021-IEEE conference on computer communications

    AdaPDP: Adaptive personalized differential privacy. In IEEE INFOCOM 2021-IEEE conference on computer communications . IEEE, 1–10

  96. [104]

    Georg Northoff and Felix Bermpohl. 2004. Cortical midline structures and the self. Trends in cognitive sciences 8, 3 (2004), 102–107. 15 Proceedings on Privacy Enhancing Technologies YYYY(X) Elfares et al

  97. [105]

    Sandra Petronio, Jeffrey T Child, and Robert D Hall. 2021. Communication privacy management theory: Significance for interpersonal communication. In Engaging theories in interpersonal communication . Routledge, 314–327

  98. [106]

    Elizabeth A Phelps and Joseph E LeDoux. 2005. Contributions of the amygdala to emotion processing: from animal models to human behavior. Neuron 48, 2 (2005), 175–187

  99. [107]

    Eric Plutzer. 2019. Privacy, sensitive questions, and informed consent: Their impacts on total survey error, and the future of survey research. Public Opinion Quarterly 83, S1 (2019), 169–184

  100. [108]

    Yong Jin Park, Scott W Campbell, and Nojin Kwak. 2012. Affect, cognition and reward: Predictors of privacy protection online. Computers in Human Behavior 28, 3 (2012), 1019–1027

  101. [109]

    John H Parmelee and Nataliya Roman. 2020. Insta-echoes: Selective exposure and selective avoidance on Instagram. Telematics and Informatics 52 (2020), 101432

  102. [110]

    Edmund T Rolls. 2004. The functions of the orbitofrontal cortex. Brain and cognition 55, 1 (2004), 11–29

  103. [111]

    Alexander K Saeri, Claudette Ogilvie, Stephen T La Macchia, Joanne R Smith, and Winnifred R Louis. 2014. Predicting Facebook users’ online privacy protection: Risk, trust, norm focus theory, and the theory of planned behavior. The Journal of social psychology 154, 4 (2014), 352–369

  104. [112]

    Pierangela Samarati and Latanya Sweeney. 1998. Protecting privacy when disclosing information: k-anonymity and its enforcement through generalization and suppression. (1998)

  105. [113]

    I} do (not) need that{Feature!

    Sarah Prange, Pascal Knierim, Gabriel Knoll, Felix Dietz, Alexander De Luca, and Florian Alt. 2024.{“I} do (not) need that{Feature!”}–Understanding{Users’} Awareness and Control of Privacy Permissions on Android Smartphones. In Twentieth Symposium on Usable Privacy and Securit...

  106. [114]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sand- hini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al

  107. [115]

    In International conference on machine learning

    Learning transferable visual models from natural language supervision. In International conference on machine learning . PmLR, 8748–8763

  108. [116]

    Malte Sönnichsen, Mayar Elfares, Yao Wang, Ralf Küsters, Alina Roitberg, and Andreas Bulling. 2025. AttentionLeak: What Does Human Attention Reveal About Information Visualisation? IACR (2025)

  109. [117]

    Julian Steil and Andreas Bulling. 2015. Discovery of everyday human activ- ities from long-term visual behaviour using topic models. In ACM UbiComp (UbiComp). ACM, 75–85. https://doi.org/10.1145/2750858.2807520

  110. [118]

    Julian Steil, Marion Koelle, Wilko Heuten, Susanne Boll, and Andreas Bulling

  111. [119]

    Negar Sammaknejad, Hamidreza Pouretemad, Changiz Eslahchi, Alireza Salahi- rad, and Ashkan Alinejad. 2017. Gender classification based on eye movements: A processing effect during passive face viewing. Advances in Cognitive Psychol- ogy 13, 3 (2017), 232

  112. [120]

    Hosnieh Sattar, Mario Fritz, and Andreas Bulling. 2020. Deep gaze pooling: Inferring and visually decoding search intents from human gaze fixations. Neu- rocomputing 387 (2020), 369–382. https://doi.org/10.1016/j.neucom.2020.01.028

  113. [121]

    Herbert A Simon. 1990. Bounded rationality. Utility and probability (1990), 15–18

  114. [122]

    Hanqiu Wang, Zihao Zhan, Haoqi Shan, Siqi Dai, Maximilian Panoff, and Shuo Wang. 2024. GAZEploit: Remote Keystroke Inference Attack by Gaze Estimation from Avatar Views in VR/MR Devices. InProceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security...

  115. [123]

    Yang Wang, Huichuan Xia, and Yun Huang. 2016. Examining American and Chinese internet users’ contextual privacy preferences of behavioral advertising. Figure 9: Demographics’ distribution of the number of male and female participants per nationality and continent In Proceeding...

  116. [124]

    good enough

    Mengxiao Zhang, Fernando Beltran, and Jiamou Liu. 2020. Selling data at an auction under privacy constraints. In Conference on Uncertainty in Artificial Intelligence. PMLR, 669–678. Appendix 1 A Eye Tracking Data 2 Here, we present a detailed breakdown of the process of data c...

  117. [126]

    Filip-Mihai Toma, Cosmin-Octavian Cepoi, Matei Nicolae Kubinschi, and Makoto Miyakoshi. 2023. Gazing through the bubble: an experimental in- vestigation into financial risk-taking using eye-tracking. Financial Innovation 9, 1 (2023), 28

  118. [127]

    Roel Vertegaal et al. 2003. Attentive user interfaces. Commun. ACM 46, 3 (2003), 30–33

  119. [128]

    Ari Ezra Waldman. 2020. Cognitive biases, dark patterns, and the ‘privacy paradox’. Current opinion in psychology 31 (2020), 105–109

  120. [132]

    We conclude that advA = 2 Pr(0|𝑆)− 1≤296 𝑒𝜀 1+𝑒𝜀− 1 = 𝑒𝜀−1 1+𝑒𝜀 .297 Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009298 20

    ⇒Pr(𝑏 = 0|𝑆) ≤𝑒𝜀 Pr(𝑏 = 1|𝑆) = 𝑒𝜀(1− Pr(𝑏 = 0|𝑆)) ⇒295 Pr(𝑏 = 0|𝑆) ≤ 𝑒𝜀 1+𝑒𝜀 . We conclude that advA = 2 Pr(0|𝑆)− 1≤296 𝑒𝜀 1+𝑒𝜀− 1 = 𝑒𝜀−1 1+𝑒𝜀 .297 Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009298 20

  121. [2019]

    In ACM ETRA

    PrivacEye: Privacy-Preserving Head-Mounted Eye Tracking Using Egocentric Scene Image and Eye Movement Features. In ACM ETRA. 1–10. https://doi.org/10.1145/3314111.3319913

  122. [2021]

    Plos one 16, 8 (2021), e0255979

    Differential privacy for eye tracking with temporal correlations. Plos one 16, 8 (2021), e0255979

  123. [2022]

    NeurIPS-GMML (2022)

    Federated Learning for Appearance-based Gaze Estimation in the Wild. NeurIPS-GMML (2022). https://doi.org/10.48550/arXiv.2211.07330

  124. [2023]

    Preprint

    PrivatEyes: Appearance-based Gaze Estimation Using Federated Secure Multi-Party Computation. Preprint. A copy has been attached to this proposal

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.