pith. sign in

arxiv: 2606.23204 · v1 · pith:SAAE7DGXnew · submitted 2026-06-22 · 💻 cs.CV

Unmasking LAION-5B: Age, Gender, Race, and Emotion Biases in Large-Scale Image Datasets

Iris Dominguez-Catena , Daniel Paternain , Mikel Galar This is my paper

Pith reviewed 2026-06-26 09:10 UTC · model grok-4.3

classification 💻 cs.CV
keywords LAION-5Bdataset biasdemographic biasage biasgender biasracial biasemotion stereotypesimage-text datasets
0
0 comments X

The pith

LAION-5B substantially overrepresents young White males while underrepresenting minority groups, older women, and certain emotion expressions.

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

The paper examines the two main components of LAION-5B by detecting faces and applying three pre-trained classifiers to assign age, gender, race, and emotion labels. It documents consistent overrepresentation of adults aged 20-39, White people, and males, with underrepresentation of other racial groups and middle-aged or older women. The same analysis uncovers stereotypical links, such as anger more often assigned to males and happiness to females. These patterns appear across both the English-only and multilingual parts of the dataset and across two different demographic classifiers. Because LAION-5B supplies training data for many generative models, the imbalances can propagate into the behavior of systems built on it.

Core claim

Using FairFace, DeepFace, and Emo-AffectNet to label faces detected in LAION-2B-en and LAION-2B-multi, the authors identify substantial overrepresentation of young adults (20-39), White individuals, and males, alongside consistent underrepresentation of minority racial groups and middle-aged or older women. They also observe stereotypical associations between demographic attributes and emotions, such as Anger being predominantly linked to males and Happiness to females. The consistency of these patterns across both dataset components and two demographic models shows that the biases are deeply embedded.

What carries the argument

Demographic and emotion labeling of detected faces by the pre-trained models FairFace, DeepFace, and Emo-AffectNet

If this is right

  • Generative models trained on LAION-5B are likely to inherit and reproduce the observed demographic imbalances in their outputs.
  • AI systems built on the dataset may show skewed performance or representation across age, gender, race, and emotion categories.
  • The documented patterns persist across both English and multilingual components, indicating the biases are not limited to one language slice.
  • Users of LAION-5B for training should incorporate explicit balancing or filtering steps to counteract the measured skews.

Where Pith is reading between the lines

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

  • The same face-labeling pipeline could be run on other large web-scraped collections to test whether comparable demographic patterns appear elsewhere.
  • If classifier error rates do vary by group, the reported bias magnitudes could shift once those rates are measured and corrected.
  • The intersectional underrepresentation of older women from minority groups may compound effects on downstream model fairness beyond the separate marginal counts.

Load-bearing premise

The error rates of the three attribute-classification models do not vary systematically with the true demographics or emotions present in the LAION images.

What would settle it

A human re-annotation of a random sample of several thousand images from each component, followed by a direct statistical comparison of the resulting age-gender-race-emotion distributions against the model outputs.

Figures

Figures reproduced from arXiv: 2606.23204 by Daniel Paternain, Iris Dominguez-Catena, Mikel Galar.

Figure 1
Figure 1. Figure 1: shows the demographic composition of LAION-2B-en (blue) and LAION-2B-multi (orange) according to both FairFace and DeepFace. 00-02 03-09 10-19 20-29 30-39 40-49 50-59 60-69 70+ FemaleMale Asian Black Indian Latino Hispanic Middle Eastern White 0.0 0.2 0.4 0.6 Proportion LAION-2B-en LAION-2B-multi (a) Demographic profile according to FairFace. 00-02 03-09 10-19 20-29 30-39 40-49 50-59 60-69 70+ FemaleMale A… view at source ↗
Figure 2
Figure 2. Figure 2: shows the facial expression distribution identified by Emo-AffectNet, dominated by “Hap￾piness” and “Neutral,” with “Fear,” “Disgust,” and “Surprise” comparatively rare. This mirrors common internet-sourced FER datasets, such as AffectNet (Mollahosseini et al., 2019; Dominguez￾Catena et al., 2024a), suggesting that LAION-5B follows broader online trends. There are minor differences between LAION-2B-en and … view at source ↗
Figure 3
Figure 3. Figure 3: Intersectional bias (Ducher’s Z) across demographic attribute pairs. Age–Race. FairFace indicates underrepresentation of the oldest groups (60+) across most races, and of very young Asian, Indian, Latino Hispanic, and Middle Eastern children; while White infants are relatively overrepresented. DeepFace shows sparser age coverage and weaker biases, but echoes the underrepresentation of older Black individua… view at source ↗
Figure 4
Figure 4. Figure 4: Emotion bias (Ducher’s Z) across demographic pairs. Emotion–Age. Stereotypical biases between emotion and age show some consistent patterns, de￾spite being weak overall. The strongest biases are the underrepresentation of age groups under 30 years in “Anger” and “Disgust”; and the underrepresentation of older groups in “Fear”, “Sadness” and “Surprise”. Emotion–Race. Effects on emotion–race are subtle and m… view at source ↗
read the original abstract

Large-scale image-text datasets, such as LAION-5B, are foundational to modern AI systems, yet their vast scale and uncurated nature raise significant concerns about demographic and stereotypical biases. This study presents a comprehensive analysis of the demographic composition and representational, stereotypical, and intersectional biases in LAION-2B-en and LAION-2B-multi, the two main components of the LAION-5B dataset. Using state-of-the-art models -- FairFace, DeepFace, and Emo-AffectNet -- we analyze faces detected in the dataset to identify biases across age, gender, race, and expressed emotion. Our findings reveal substantial overrepresentation of young adults (20--39), White individuals, and males, alongside consistent underrepresentation of minority racial groups and middle-aged or older women across both dataset components. We also observe stereotypical associations between demographic attributes and emotions, such as ``Anger'' being predominantly linked to males and ``Happiness'' to females, pointing to systemic imbalances in the data. The consistency of these patterns across two demographic models and both components of LAION-5B demonstrates that these biases are deeply embedded in one of the most widely-used training datasets. Given the scale at which LAION-5B is used to train generative models, these demographic imbalances could shape the behavior and outputs of numerous downstream AI systems.

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 / 1 minor

Summary. The paper analyzes demographic (age, gender, race) and emotion biases in the LAION-2B-en and LAION-2B-multi components of LAION-5B by applying face detection followed by the classifiers FairFace, DeepFace, and Emo-AffectNet. It reports substantial overrepresentation of young adults (20-39), White individuals, and males, underrepresentation of minority racial groups and middle-aged/older women, and stereotypical emotion associations (e.g., Anger predominantly with males, Happiness with females), with consistency across the two dataset components and two demographic models.

Significance. If the classifier-derived counts accurately reflect image content, the work provides a valuable empirical baseline on biases in one of the largest publicly used image-text datasets, with direct relevance to fairness in generative models and downstream CV systems. The reported consistency across components and models is a positive empirical feature.

major comments (2)
  1. [§3 (Methodology)] §3 (Methodology): No stratified validation, calibration curves, or human-labeled accuracy metrics are reported for FairFace, DeepFace, or Emo-AffectNet on LAION images or a comparable held-out set; without this, differential error rates by age/gender/race cannot be ruled out as a source of the reported imbalances.
  2. [§4 (Results)] §4 (Results) and Abstract: The central percentages and association claims rest on the untested assumption that classifier error rates do not covary with the true demographics or emotions; no sensitivity analysis or error-propagation bounds are supplied to quantify how plausible misclassification patterns would alter the over/under-representation findings.
minor comments (1)
  1. The abstract would be clearer if it stated the total number of detected faces, the face-detection threshold used, and the fraction of images discarded due to no-face or low-confidence detections.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on classifier validation and robustness. We address each major comment below and will revise the manuscript accordingly to improve methodological transparency.

read point-by-point responses

  1. Referee: [§3 (Methodology)] §3 (Methodology): No stratified validation, calibration curves, or human-labeled accuracy metrics are reported for FairFace, DeepFace, or Emo-AffectNet on LAION images or a comparable held-out set; without this, differential error rates by age/gender/race cannot be ruled out as a source of the reported imbalances.

    Authors: The manuscript selects FairFace, DeepFace, and Emo-AffectNet as established, publicly documented tools for demographic and emotion inference, relying on the performance metrics reported in their original publications rather than re-validating them on LAION. No new stratified validation, calibration curves, or human-labeled metrics on LAION (or a comparable set) are provided because the study focus is an audit of dataset composition, not a benchmark of the classifiers themselves. We acknowledge that this leaves open the possibility of differential error rates contributing to observed imbalances. In revision we will add an expanded discussion in §3 of the classifiers' documented limitations and a new limitations subsection that explicitly flags the absence of LAION-specific validation as a constraint on interpreting the counts. revision: yes

  2. Referee: [§4 (Results)] §4 (Results) and Abstract: The central percentages and association claims rest on the untested assumption that classifier error rates do not covary with the true demographics or emotions; no sensitivity analysis or error-propagation bounds are supplied to quantify how plausible misclassification patterns would alter the over/under-representation findings.

    Authors: We agree that the reported percentages and associations would be strengthened by explicit quantification of sensitivity to plausible misclassification. While the manuscript already notes consistency of patterns across two independent demographic classifiers, this does not constitute a formal sensitivity analysis. In the revised version we will insert a new subsection in §4 that performs a sensitivity analysis: we will simulate misclassification matrices drawn from the published error rates of FairFace, DeepFace, and Emo-AffectNet, propagate these errors through the demographic and emotion distributions, and report bounds on how the over- and under-representation statistics could shift under different error-covariance assumptions. Corresponding caveats will be added to the abstract. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical counts from external classifiers on external data

full rationale

The paper performs a direct empirical audit: it runs three pre-trained, externally developed classifiers (FairFace, DeepFace, Emo-AffectNet) on faces detected in the public LAION-5B subsets and tabulates the resulting label distributions. No equations, fitted parameters, or self-citations are used to derive the reported over/under-representations or emotion associations; the counts are literal outputs of the external models. The work therefore contains no self-definitional steps, no fitted-input predictions, and no load-bearing self-citation chains. The skeptic concern about classifier error rates varying by demographic is a validity question, not a circularity question under the stated criteria.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified accuracy of three external face-attribute models and on the assumption that face detection itself does not introduce additional demographic selection bias.

axioms (1)
  • domain assumption FairFace, DeepFace, and Emo-AffectNet labels are sufficiently accurate and unbiased for the purpose of measuring dataset composition.
    The paper treats the outputs of these models as ground truth for age, gender, race, and emotion without reporting error rates or calibration on LAION images.

pith-pipeline@v0.9.1-grok · 5791 in / 1250 out tokens · 19306 ms · 2026-06-26T09:10:05.610348+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

297 extracted references · 207 canonical work pages · 12 internal anchors

  1. [1]

    2021 , month = jan, institution =

    Emotional. 2021 , month = jan, institution =

    2021

  2. [2]

    2019 , month = may, journal =

    Fairness in Representation: Quantifying Stereotyping as a Representational Harm , author =. 2019 , month = may, journal =. doi:10.1137/1.9781611975673 , urldate =

    work page doi:10.1137/1.9781611975673 2019

  3. [3]

    and Pasquetto, I.V

    Abdu, Amina A. and Pasquetto, Irene V. and Jacobs, Abigail Z. , year =. An. 2023. doi:10.1145/3593013.3594083 , urldate =

    work page doi:10.1145/3593013.3594083 2023

  4. [4]

    2005 , month = jan, publisher =

    Computer. 2005 , month = jan, publisher =

    2005

  5. [5]

    2013 , month = jan, publisher =

    Computer. 2013 , month = jan, publisher =. doi:10.1145/2534860 , urldate =

    work page doi:10.1145/2534860 2013

  6. [6]

    2023 , journal =

    Directly. 2023 , journal =. doi:10.1111/1468-2230.12759 , urldate =

    work page doi:10.1111/1468-2230.12759 2023

  7. [7]

    Agarwal, Alekh and Dud. Fair. 2019 , month = may, journal =. doi:10.48550/arXiv.1905.12843 , urldate =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1905.12843 2019

  8. [8]

    Comparing the

    Ahmad, Khurshid and Wang, Shirui and Vogel, Carl and Jain, Pranav and O'Neill, Oscar and Sufi, Basit Hamid , editor =. Comparing the. Proceedings of the. 2022 , volume =. doi:10.1007/978-3-030-89906-6\_14 , urldate =

    work page doi:10.1007/978-3-030-89906-6 2022

  9. [9]

    , year =

    Aifanti, Niki and Papachristou, Christos and Delopoulos, A. , year =. The. 11th Int. Workshop Image Anal. Multimed. Interact. Serv. WIAMIS 10 , urldate =

  10. [10]

    arXiv , langid =:2402.01002 , primaryclass =

    AlDahoul, Nouar and Rahwan, Talal and Zaki, Yasir , year =. arXiv , langid =:2402.01002 , primaryclass =

    arXiv

  11. [11]

    Modeling

    Aneja, Deepali and Colburn, Alex and Faigin, Gary and Shapiro, Linda and Mones, Barbara , editor =. Modeling. Comput. 2017 , series =. doi:10.1007/978-3-319-54184-6\_9 , abstract =

    work page doi:10.1007/978-3-319-54184-6 2017

  12. [12]

    Angwin, Julia and Larson, Jeff and Mattu, Surya and Kirchner, Lauren , year =. Machine. ProPublica , urldate =

  13. [13]

    and Busby, Ethan and Gubler, Joshua and Bail, Chris and Howe, Thomas and Rytting, Christopher and Wingate, David , year =

    Argyle, Lisa P. and Busby, Ethan and Gubler, Joshua and Bail, Chris and Howe, Thomas and Rytting, Christopher and Wingate, David , year =. arXiv , langid =:2302.07268 , primaryclass =

    arXiv

  14. [14]

    Arp, Daniel and Quiring, Erwin and Pendlebury, Feargus and Warnecke, Alexander and Pierazzi, Fabio and Wressnegger, Christian and Cavallaro, Lorenzo and Rieck, Konrad , year =. Dos and. arXiv , file =:2010.09470 , primaryclass =

    arXiv 2010

  15. [15]

    Assuncao, Gustavo and Patrao, Bruno and. An. 2022 , journal =. doi:10.1109/TAI.2022.3159614 , abstract =

    work page doi:10.1109/tai.2022.3159614 2022

  16. [16]

    2005 , month = oct, journal =

    Asymmetric Interference between Sex and Emotion in Face Perception , author =. 2005 , month = oct, journal =. doi:10.3758/BF03193553 , urldate =

    work page doi:10.3758/bf03193553 2005

  17. [17]

    Avella, Marcela del Pilar Roa , year =. Crime. Telecommun. Syst. Manag. , volume =

  18. [18]

    , year =

    Bagert, Donald J. , year =. Viewpoint: Taking the Lead in Licensing Software Engineers , shorttitle =. Commun. ACM , volume =. doi:10.1145/299157.299163 , urldate =

    work page doi:10.1145/299157.299163

  19. [19]

    Constitutional AI: Harmlessness from AI Feedback

    Bai, Yuntao and Kadavath, Saurav and Kundu, Sandipan and Askell, Amanda and Kernion, Jackson and Jones, Andy and Chen, Anna and Goldie, Anna and Mirhoseini, Azalia and McKinnon, Cameron and Chen, Carol and Olsson, Catherine and Olah, Christopher and Hernandez, Danny and Drain, Dawn and Ganguli, Deep and Li, Dustin and. Constitutional. 2022 , month = dec, ...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2212.08073 2022

  20. [20]

    Facial Emotion Recognition and Music Recommendation System Using

    Bakariya, Brijesh and Singh, Arshdeep and Singh, Harmanpreet and Raju, Pankaj and Rajpoot, Rohit and Mohbey, Krishna Kumar , year =. Facial Emotion Recognition and Music Recommendation System Using. Evolving Systems , issn =. doi:10.1007/s12530-023-09506-z , urldate =

    work page doi:10.1007/s12530-023-09506-z

  21. [21]

    Introducing the

    B. Introducing the. 2012 , journal =. doi:10.1037/a0025827 , urldate =

    work page doi:10.1037/a0025827 2012

  22. [22]

    2019 , publisher =

    Fairness and Machine Learning , author =. 2019 , publisher =

    2019

  23. [23]

    Barrett, Teanna and Chen, Quanze and Zhang, Amy , year =. Skin. Proc. 2023. doi:10.1145/3593013.3594114 , urldate =

    work page doi:10.1145/3593013.3594114 2023

  24. [24]

    In: Proceedings of the 18th ACM International Conference on Multimodal Interactio

    Barsoum, Emad and Zhang, Cha and Ferrer, Cristian Canton and Zhang, Zhengyou , year =. Training Deep Networks for Facial Expression Recognition with Crowd-Sourced Label Distribution , booktitle =. doi:10.1145/2993148.2993165 , urldate =

    work page doi:10.1145/2993148.2993165

  25. [25]

    and Kassler, Andreas , year =

    Bayram, Firas and Ahmed, Bestoun S. and Kassler, Andreas , year =. From. doi:10.48550/arXiv.2203.11070 , urldate =. arXiv , keywords =:2203.11070 , primaryclass =

    work page doi:10.48550/arxiv.2203.11070

  26. [26]

    Beaupr. Cross-. 2005 , month = may, journal =. doi:10.1177/0022022104273656 , urldate =

    work page doi:10.1177/0022022104273656 2005

  27. [27]

    2007 , journal =

    The Confounded Nature of Angry Men and Happy Women , author =. 2007 , journal =. doi:10.1037/0022-3514.92.2.179 , abstract =

    work page doi:10.1037/0022-3514.92.2.179 2007

  28. [28]

    2024 , month = nov, journal =

    Ethics Dumping in Artificial Intelligence , author =. 2024 , month = nov, journal =. doi:10.3389/frai.2024.1426761 , urldate =

    work page doi:10.3389/frai.2024.1426761 2024

  29. [29]

    1966 , address =

    A Comparison of Some Cluster-Seeking Techniques , author =. 1966 , address =

    1966

  30. [30]

    doi:10.1145/3442188.3445924 , isbn =

    Bender, Emily M. and Gebru, Timnit and. On the. Proc. 2021. 2021 , month = mar, series =. doi:10.1145/3442188.3445922 , urldate =

    work page doi:10.1145/3442188.3445922 2021

  31. [31]

    2016 , month = jun, pages =

    2016. 2016 , month = jun, pages =. doi:10.1109/CVPR.2016.600 , urldate =

    work page doi:10.1109/cvpr.2016.600 2016

  32. [32]

    and Parker, Frances L

    Berger, Wolfgang H. and Parker, Frances L. , year =. Diversity of. Science , volume =. doi:10.1126/science.168.3937.1345 , urldate =

    work page doi:10.1126/science.168.3937.1345

  33. [33]

    A Bias-Correction for

    Bergsma, Wicher , year =. A Bias-Correction for. Journal of the Korean Statistical Society , volume =. doi:10.1016/j.jkss.2012.10.002 , urldate =

    work page doi:10.1016/j.jkss.2012.10.002 2012

  34. [34]

    and Heidari, Hoda and Jabbari, S

    Berk, R. and Heidari, Hoda and Jabbari, S. and Joseph, Matthew and Kearns, Michael and Morgenstern, Jamie and Neel, Seth and Roth, Aaron , year =. A. ArXiv , doi =

  35. [35]

    Fairness in

    Berk, Richard and Heidari, Hoda and Jabbari, Shahin and Kearns, Michael and Roth, Aaron , year =. Fairness in. Sociological Methods & Research , volume =. doi:10.1177/0049124118782533 , urldate =

    work page doi:10.1177/0049124118782533

  36. [36]

    2022 , month = jun, journal =

    Potential Sources of Dataset Bias Complicate Investigation of Underdiagnosis by Machine Learning Algorithms , author =. 2022 , month = jun, journal =. doi:10.1038/s41591-022-01846-8 , urldate =

    work page doi:10.1038/s41591-022-01846-8 2022

  37. [37]

    Are We Done with

    Beyer, Lucas and H. Are We Done with. 2020 , month = jun, journal =. arXiv , langid =:2006.07159 , primaryclass =

    arXiv 2020

  38. [38]

    Matsumoto and

    Biehl, Michael and Matsumoto, David and Ekman, Paul and Hearn, Valerie and Heider, Karl and Kudoh, Tsutomu and Ton, Veronica , year =. Matsumoto and. Journal of Nonverbal Behavior , volume =. doi:10.1023/A:1024902500935 , urldate =

    work page doi:10.1023/a:1024902500935

  39. [39]

    Multimodal Datasets: Misogyny, Pornography, and Malignant Stereotypes , shorttitle =

    Birhane, Abeba and Prabhu, Vinay Uday and Kahembwe, Emmanuel , year =. Multimodal Datasets: Misogyny, Pornography, and Malignant Stereotypes , shorttitle =. ArXiv211001963 Cs , eprint =

  40. [40]

    In: IEEE Winter Conference on Applications of Computer Vision (WACV)

    Birhane, Abeba and Prabhu, Vinay Uday , year =. Large Image Datasets:. 2021. doi:10.1109/WACV48630.2021.00158 , urldate =

    work page doi:10.1109/wacv48630.2021.00158 2021

  41. [41]

    Birhane, Abeba and Prabhu, Vinay and Han, Sang and Boddeti, Vishnu Naresh , year =. On. ArXiv230613141 Cs , doi =

  42. [42]

    Birhane, Abeba and Dehdashtian, Sepehr and Prabhu, Vinay and Boddeti, Vishnu , year =. The. Proc. 2024. doi:10.1145/3630106.3658968 , urldate =

    work page doi:10.1145/3630106.3658968 2024

  43. [43]

    2017 , month = oct, journal =

    The Visibility of Social Class from Facial Cues , author =. 2017 , month = oct, journal =. doi:10.1037/pspa0000091 , abstract =

    work page doi:10.1037/pspa0000091 2017

  44. [44]

    Measuring

    Blakeney, Cody and Atkinson, Gentry and Huish, Nathaniel and Yan, Yan and Metsis, Vangelis and Zong, Ziliang , year =. Measuring. 2022. doi:10.1109/NAS55553.2022.9925287 , urldate =

    work page doi:10.1109/nas55553.2022.9925287 2022

  45. [45]

    Learning to Teach in Higher Education:

    Bligh, Donald , year =. Learning to Teach in Higher Education:. Studies in Higher Education , volume =. doi:10.1080/03075079312331382498 , urldate =

    work page doi:10.1080/03075079312331382498

  46. [46]

    Language (Technology) is Power: A Critical Survey of ``Bias'' in NLP

    Blodgett, Su Lin and Barocas, Solon and Daum. Language (. Proc. 58th. 2020 , pages =. doi:10.18653/v1/2020.acl-main.485 , urldate =

    work page doi:10.18653/v1/2020.acl-main.485 2020

  47. [47]

    On the Opportunities and Risks of Foundation Models

    Bommasani, Rishi and Hudson, Drew A. and Adeli, Ehsan and Altman, Russ and Arora, Simran and von Arx, Sydney and Bernstein, Michael S. and Bohg, Jeannette and Bosselut, Antoine and Brunskill, Emma and Brynjolfsson, Erik and Buch, Shyamal and Card, Dallas and Castellon, Rodrigo and Chatterji, Niladri and Chen, Annie and Creel, Kathleen and Davis, Jared Qui...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2108.07258 2022

  48. [48]

    2016 , month = nov, journal =

    Stereotypes* , author =. 2016 , month = nov, journal =. doi:10.1093/qje/qjw029 , urldate =

    work page doi:10.1093/qje/qjw029 2016

  49. [49]

    Normalized (Pointwise) Mutual Information in Collocation Extraction , booktitle =

    Bouma, Gerlof , year =. Normalized (Pointwise) Mutual Information in Collocation Extraction , booktitle =

  50. [50]

    , year =

    Brock, David A. , year =. Comparison of. J. Water Pollut. Control Fed. , volume =. 25039481 , eprinttype =

  51. [51]

    and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, J

    Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, J. and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and. Language. 2020 , month = may, journal =

    2020

  52. [52]

    Gender Shades:

    Buolamwini, Joy and Gebru, Timnit , editor =. Gender Shades:. Proc. 1st. 2018 , month = feb, series =

    2018

  53. [53]

    , year =

    Burkart, Nadia and Huber, Marco F. , year =. A. J. Artif. Intell. Res. , volume =. doi:10.1613/jair.1.12228 , urldate =

    work page doi:10.1613/jair.1.12228

  54. [54]

    and Lee, Sungbok and Narayanan, Shrikanth S

    Busso, Carlos and Bulut, Murtaza and Lee, Chi-Chun and Kazemzadeh, Abe and Mower, Emily and Kim, Samuel and Chang, Jeannette N. and Lee, Sungbok and Narayanan, Shrikanth S. , year =. Lang Resources & Evaluation , volume =. doi:10.1007/s10579-008-9076-6 , urldate =

    work page doi:10.1007/s10579-008-9076-6

  55. [55]

    Buyl, Maarten and Rogiers, Alexander and Noels, Sander and. Large. 2024 , month = oct, number =. arXiv , keywords =:2410.18417 , journal =

    arXiv 2024

  56. [56]

    Building

    Calders, Toon and Kamiran, Faisal and Pechenizkiy, Mykola , year =. Building. 2009. doi:10.1109/ICDMW.2009.83 , abstract =

    work page doi:10.1109/icdmw.2009.83 2009

  57. [57]

    Cannarsa, Michel , editor =. Ethics. The. 2021 , series =. doi:10.1017/9781108936040.022 , urldate =

    work page doi:10.1017/9781108936040.022 2021

  58. [58]

    and Keutmann, Michael K

    Cao, Houwei and Cooper, David G. and Keutmann, Michael K. and Gur, Ruben C. and Nenkova, Ani and Verma, Ragini , year =. IEEE Trans Affect Comput , volume =. doi:10.1109/TAFFC.2014.2336244 , urldate =

    work page doi:10.1109/taffc.2014.2336244 2014

  59. [59]

    and Zisserman, Andrew , year =

    Cao, Qiong and Shen, Li and Xie, Weidi and Parkhi, Omkar M. and Zisserman, Andrew , year =. arXiv , langid =:1710.08092 , primaryclass =

    Pith/arXiv arXiv

  60. [60]

    2024 , month = may, journal =

    Casta. 2024 , month = may, journal =. doi:10.1016/j.softx.2024.101728 , urldate =

    work page doi:10.1016/j.softx.2024.101728 2024

  61. [61]

    2020 , month = nov, publisher =

    Computing. 2020 , month = nov, publisher =. doi:10.1145/3467967 , urldate =

    work page doi:10.1145/3467967 2020

  62. [62]

    Llama Meets

    Chalkidis, Ilias and Brandl, Stephanie , year =. Llama Meets. arXiv , langid =:2403.13592 , primaryclass =

    arXiv

  63. [63]

    and Cramer, Mackenzie and Soni, Sandeep and Bamman, David , year =

    Chang, Kent K. and Cramer, Mackenzie and Soni, Sandeep and Bamman, David , year =. Speak,. ArXiv230500118 Cs , eprint =

  64. [64]

    Chang, Hongyan and Shokri, Reza , year =. Bias. doi:10.48550/ARXIV.2309.02160 , urldate =

    work page doi:10.48550/arxiv.2309.02160

  65. [65]

    Practical

    Chaudhary, Bhushan and Pandey, Anubha and Bhatt, Deepak and Tiwari, Darshika , year =. Practical. doi:10.48550/arXiv.2312.15994 , urldate =. arXiv , keywords =:2312.15994 , publisher =

    work page doi:10.48550/arxiv.2312.15994

  66. [66]

    2017 , month = feb, journal =

    Smile Detection in the Wild with Deep Convolutional Neural Networks , author =. 2017 , month = feb, journal =. doi:10.1007/s00138-016-0817-z , urldate =

    work page doi:10.1007/s00138-016-0817-z 2017

  67. [67]

    2018 , month = oct, journal =

    Distinct Facial Expressions Represent Pain and Pleasure across Cultures , author =. 2018 , month = oct, journal =. doi:10.1073/pnas.1807862115 , urldate =

    work page doi:10.1073/pnas.1807862115 2018

  68. [68]

    Selbst, Danah Boyd, Sorelle A

    Chen, Jiahao and Kallus, Nathan and Mao, Xiaojie and Svacha, Geoffry and Udell, Madeleine , year =. Fairness. Proc. doi:10.1145/3287560.3287594 , urldate =. arXiv , langid =:1811.11154 , primaryclass =

    work page doi:10.1145/3287560.3287594

  69. [69]

    De Lange, T

    Chen, Yunliang and Joo, Jungseock , year =. Understanding and. 2021. doi:10.1109/ICCV48922.2021.01471 , abstract =

    work page doi:10.1109/iccv48922.2021.01471 2021

  70. [70]

    Cheong, Jiaee and Kalkan, Sinan and Gunes, Hatice , year =. The. IEEE Signal Process. Mag. , volume =. doi:10.1109/MSP.2021.3106619 , abstract =

    work page doi:10.1109/msp.2021.3106619 2021

  71. [71]

    Investigating

    Cheong, Marc and Abedin, Ehsan and Ferreira, Marinus and Reimann, Ritsaart and Chalson, Shalom and Robinson, Pamela and Byrne, Joanne and Ruppanner, Leah and Alfano, Mark and Klein, Colin , year =. Investigating. ACM J. Responsib. Comput. , volume =. doi:10.1145/3649883 , urldate =

    work page doi:10.1145/3649883

  72. [72]

    doi:10.48550/arXiv.2407.18745 , urldate =

    Chinta, Sribala Vidyadhari and Wang, Zichong and Yin, Zhipeng and Hoang, Nhat and Gonzalez, Matthew and Quy, Tai Le and Zhang, Wenbin , year =. doi:10.48550/arXiv.2407.18745 , urldate =. arXiv , file =:2407.18745 , primaryclass =

    work page doi:10.48550/arxiv.2407.18745

  73. [73]

    2020 , month = apr, journal =

    A Snapshot of the Frontiers of Fairness in Machine Learning , author =. 2020 , month = apr, journal =. doi:10.1145/3376898 , urldate =

    work page doi:10.1145/3376898 2020

  74. [74]

    Christoforaki, Maria and Beyan, Oya , year =. Appl. Sci. , volume =. doi:10.3390/app12094130 , urldate =

    work page doi:10.3390/app12094130

  75. [75]

    Churamani, Nikhil and Kara, Ozgur and Gunes, Hatice , year =. Domain-. IEEE Trans. Affective Comput. , eprint =. doi:10.1109/TAFFC.2022.3181033 , urldate =

    work page doi:10.1109/taffc.2022.3181033 2022

  76. [76]

    and Rubin, Donald B

    Cochran, William G. and Rubin, Donald B. , year =. Controlling. Sankhy. 25049893 , eprinttype =

  77. [78]

    Statistical Power Analysis for the Behavioral Sciences , ISBN =

    Cohen, Jacob , year =. Statistical. doi:10.4324/9780203771587 , urldate =

    work page doi:10.4324/9780203771587

  78. [79]

    1987 , address =

    Informe:. 1987 , address =

    1987

  79. [80]

    1988 , address =

    Informe:. 1988 , address =

    1988

  80. [81]

    Reflexive

    Cooper, A. Feder and Abrams, Ellen , year =. Emergent. Proc. 2021 AAAIACM Conf. AI Ethics Soc. , eprint =. doi:10.1145/3461702.3462519 , urldate =

    work page doi:10.1145/3461702.3462519 2021

Showing first 80 references.