Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Addressing Bias in LLMs: Strategies and Application to Fair AI-based Recruitment

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A privacy-enhancing framework that removes gender information from either the input text or the hidden representations can stop Transformer-based resume scorers from reproducing gender bias in their training data.

desk verdict Useful case study with a real evaluation leak: Approach 1's token selection and evaluation share the same validation split, so the paper's strongest result is not trustworthy as reported. read the letter →

arxiv 2506.11880 v1 pith:C2ZHJHWI submitted 2025-06-13 cs.AI cs.CL

Alejandro Peña , Julian Fierrez , Aythami Morales , Gonzalo Mancera , Miguel Lopez , Ruben Tolosana This is my paper
classification cs.AIcs.CL
keywords genderbiaslargelanguagemodelsmitigationautomatedrecruitmentalgorithmicfairnessIntegratedGradientsadversariallearningFairCVdb
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

This paper argues that Transformer-based resume scoring systems will learn gender bias whenever the training scores are biased by gender, and that the bias can be stopped by keeping gender information out of the learning pipeline. On the FairCVdb testbed of 24,000 synthetic profiles, BERT- and RoBERTa-based scorers trained on gender-biased scores produced shortlists that were about two-thirds male and fell below the U.S. 4/5 fairness rule. Two interventions — masking gender-linked words found by Integrated Gradients, and an adversarial learning-not-to-learn step that strips gender from hidden representations — both restored near-equal demographic ratios while improving overall recall. The result matters because biased shortlisting is an allocational harm: it withholds job opportunities from one demographic group, and the proposed framework prevents that even when the training labels themselves are biased.

What carries the argument

The load-bearing machinery is a two-stage pipeline — a frozen Transformer text encoder (BERT or RoBERTa) followed by a small MLP fusion head — plus two interventions that cut gender information from different stages. Approach 1 applies Integrated Gradients to attribute each token's contribution to the predicted score, collects the most relevant tokens per gender and labor sector, and masks gender-linked and stereotype-linked words before retraining. Approach 2 adapts learning-not-to-learn (LNTL): an auxiliary classifier tries to predict gender from a 300-dimensional hidden layer of the fusion head, and a min-max adversarial objective minimises the mutual information between that layer and gender while preserving score prediction. Both interventions target the channel that lets the model connect text to the biased target: the latent representation of the biography.

What would settle it

A direct falsifier would be to run both mitigation methods on a real resume corpus with naturally occurring gender-annotated hiring outcomes: if the demographic ratio of the shortlist stays below 0.8 or the score-distribution divergence stays above roughly 0.05 after either intervention, the claim that these methods prevent biased behaviour does not transfer beyond the synthetic testbed.

Watch

Extended reading notes

Core claim

The central discovery is that a recruitment model can be prevented from reproducing data bias even when the training labels are biased. When the scoring target is gender-biased, both BERT- and RoBERTa-based scorers encode gender in their latent text representations and use it to inflate male candidates' scores, dropping the shortlist demographic ratio to about 0.44-0.49. Removing gender information in the input space (replacing gender-linked tokens and names with [MASK]) brings the ratio above 0.96; removing it in latent space via adversarial training brings the ratio above 0.83 and yields the highest overall recall of all configurations. The adversarial method still leaves a residual gender signal visible in t-SNE projections, but the system satisfies the 4/5 fairness rule and no longer reproduces the biased labels.

Load-bearing premise

The load-bearing premise is that the synthetic FairCVdb data, whose biased scores are created by artificially injecting gender into a weighted formula, behaves like real resumes with naturally occurring annotation bias.

Editorial extensions

If this is right

  • When trained on biased scores, both BERT- and RoBERTa-based scorers show demographic ratios around 0.44-0.49, below the 4/5 rule threshold, confirming that Transformer models exploit gender proxies in text.
  • Masking gender-linked tokens detected by Integrated Gradients brings the demographic ratio above 0.96 and the score-distribution divergence below 0.03, while improving overall recall over the biased baseline.
  • Adversarial removal of gender from the hidden layer satisfies the 4/5 rule with ratios above 0.83 and yields the highest recall values (83.0% for BERT, 80.0% for RoBERTa) of any configuration.
  • Neither method requires changing the pretrained language model; both operate on the input text or on the small fusion head, so they can be added to existing resume-scoring pipelines.
  • The biased baseline's latent embeddings cluster by gender rather than by labor sector; after adversarial training, labor-sector structure returns but a within-cluster gender distinction remains.

Reading between the lines

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

  • A natural testable extension is to combine both methods: mask gender proxies first and apply LNTL afterward, which might remove the residual within-cluster gender separation seen in t-SNE and push the latent-space configuration closer to the near-perfect parity of Approach 1.
  • The same input-space and latent-space interventions should transfer to other sensitive attributes such as ethnicity (FairCVdb already includes ethnicity-biased scores) and to other high-stakes text tasks like loan or admissions screening, though each new context will need its own proxy-token discovery.
  • Because FairCVdb's biographies come from a public dataset of occupation descriptions, the list of proxy tokens (e.g., 'children', 'family', 'husband') is likely domain-specific; in real resumes the proxies may shift, so an automated token-discovery step is essential rather than a fixed word list.
  • The paper treats gender as binary because the dataset does; the method itself is attribute-agnostic, but applying it to non-binary or intersectional categories would require data and evaluation criteria the current testbed does not provide.
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 addresses gender bias in Transformer-based resume scoring systems. It proposes and evaluates two bias mitigation methods on the FairCVdb synthetic dataset: Approach 1, which uses Integrated Gradients to identify gender-proxy tokens in biographies and masks them before retraining; and Approach 2, an adaptation of Learning-Not-to-Learn (LNTL) that adversarially removes gender information from a latent representation. Experiments with BERT and RoBERTa show that biased training labels lead to biased predictions, and that both proposed methods reduce gender imbalance in a top-500 shortlist according to demographic ratio, D_KL, and recall metrics. The paper concludes that both methods successfully reduced gender information and prevented biased behaviors despite biased training data.

Significance. If the results are valid, the paper offers a practical comparison of two debiasing strategies for a high-stakes NLP application, with the explicit goal of reducing allocational harm in AI-based recruitment. The work is positioned within an active and important research area, and the use of a public synthetic testbed (FairCVdb) makes the experiments reproducible in principle. Strengths include the evaluation across two transformer architectures, multiple fairness criteria (statistical parity, equality of opportunity, 4/5 rule), and a qualitative t-SNE analysis that connects latent geometry to the bias-removal mechanism. However, the validity of the central claim hinges on the integrity of the evaluation protocol, and the synthetic nature of the dataset limits external generalization. The paper also provides no statistical significance testing or error bars, so the quantitative differences between methods are not yet firmly established.

major comments (4)
  1. [Approach 1: Via Model Explainability] The evaluation protocol for Approach 1 suffers from a form of evaluation leakage: the same validation split is used both to select the gender-proxy tokens via Integrated Gradients and to compute the fairness metrics in Table 2. The text states that attributions are computed on the validation set, the top tokens aggregated by gender and sector, and the resulting mask is applied before retraining; all reported proportions, ratios, D_KL, and recall values in Table 2 are then measured on that same validation split. This can inflate the apparent debiasing because the mask is fitted to the very split used for evaluation. A held-out test set, or a nested/cross-validated protocol, is required to support the claim that Approach 1 'successfully reduced gender information' in a generalizable way. This issue is load-bearing because the strongest version of the central conclusion ('both methods successfully reduced gender information') depends on the Approach 1 numbers in Table 2.
  2. [Table 2 / Approach 2: Via Adversarial Learning] Approach 2 leaves measurable residual bias: the demographic ratio is 0.845 (BERT) and 0.831 (RoBERTa), D_KL is 0.0422 and 0.0487, and the recall gap between men and women is about 6.8 percentage points for BERT (86.4% vs 79.6%). The authors acknowledge in the text that 'gender information was not completely removed,' yet the Conclusions state that both methods 'successfully reduced gender information in the pipeline, thus preventing biased behaviors.' This overstates the evidence. The paper should either soften the conclusion to 'substantially reduced but not eliminated' or provide a formal statistical test showing that the residual gap is not significant. Without such a test, the claims about 'preventing biased behaviors' are not supported by the reported effect sizes.
  3. [Materials and Methods: Dataset: FairCVdb] The entire evaluation rests on FairCVdb, a synthetic dataset of 24,000 profiles whose ground-truth scores are generated by a weighted linear combination of competencies and whose biased scores are obtained by artificially injecting gender-based bias. The paper provides no validation on real resumes or real annotation processes. If real hiring pipelines exhibit different bias mechanisms (e.g., bias in free-text descriptions, contextual cues, or label noise beyond a simple additive gender shift), the reported mitigation results may not transfer. The authors should discuss this limitation explicitly and, if possible, include at least a small real-world case study or a more realistic bias injection protocol to strengthen external validity.
  4. [Experimental Setup and Results / Table 2] All results in Tables 1 and 2 are point estimates with no variability measures (e.g., standard deviations or confidence intervals across random seeds). Since the comparison between Approach 1 and Approach 2 relies on small differences in D_KL (e.g., 0.0267 vs 0.0422 for BERT), and since training runs may vary with initialization, the paper should report repeated-run statistics or at least error bars. Without these, it is not possible to determine whether the observed differences are meaningful or within noise.
minor comments (5)
  1. [Abstract] There is a typo: 'they are are susceptible' should be 'they are susceptible.'
  2. [Fairness Criteria] The phrase 'prima facia evidence' should be 'prima facie evidence.'
  3. [Throughout] Several cross-references are left as empty 'Sect.' and 'Section' placeholders (e.g., in 'Experimental Setup and Results' and 'Approach 1'). These should be resolved to actual section numbers before publication.
  4. [Approach 2: Via Adversarial Learning] The sentence 'the tool clearly has some sense of it' is informal for a journal and would benefit from a more precise statement, e.g., 'the latent representations still contain some gender-discriminative information, as evidenced by the within-cluster separation in the t-SNE projection.'
  5. [Conclusions] The conclusion that 'both methods successfully reduced gender information' should explicitly acknowledge the residual bias in Approach 2 and the evaluation-leakage risk in Approach 1, since these are directly relevant to the strength of the claims.

Circularity Check

1 steps flagged · score 6.0 of 10

Approach 1's bias-reduction result is an in-sample statistic: the Integrated Gradients token mask is chosen on the validation split and Table 2's fairness metrics are computed on the same split, so the headline D_KL drop is partly a fitted result rather than an unbiased estimate.

  1. fitted input called prediction [Approach 1: Via Model Explainability; Table 2; Experimental Setup and Results (train/validation splits from FairCVtest)]
    "This technique measures token relevance in a validation set and detect relevant words ... For each resume, we extract the 20 most relevant tokens. We then extract the frequency of the tokens in this set of tokens by gender and labor sector, obtaining for each of the resulting 8 groups a set of the 30 most relevant tokens. ... We replace the previous tokens, as well as the bios names, with the BERT/RoBERTa mask token ([MASK]), and retrain the scoring tool W_S with the biased scores. ... Table 2: Gender proportions in the top-500 candidates of the validation set of FairCVtest."

    The token mask is fitted to the evaluation split: IG attributions are computed on validation resumes, tokens are aggregated by gender and sector, and the top-30 per group are masked; Table 2 then reports fairness metrics on the same validation split. The D_KL drop from about 0.31/0.34 to about 0.02 is measured on the very set whose gender labels were used to choose which gender-correlated tokens to remove. This is an in-sample statistic, not an unbiased estimate on unseen resumes; the paper reports no separate test set. The conclusion that Approach 1 'successfully reduced gender information' is thus partly supported by fitting the mask to the evaluation set. Approach 2 is not affected, so the 'both methods' claim retains some independent evidence.

full rationale

The paper is not circular in its mathematical derivation: the LNTL objective (Eq. 1) comes from Kim et al. (2019), the task is a standard regression with an auxiliary adversarial classifier, and the baseline analysis (Table 1) is a direct demonstration that BERT/RoBERTa learn the artificially injected gender bias in FairCVdb. Self-citations to FairCVdb/FairCVtest and to prior LNTL applications (Pena et al. 2021; Morales et al. 2021) are used as provenance for the dataset and the adaptation, not as an unverified uniqueness theorem or as a definition of the target result, so they are not load-bearing circularity. The central circularity concern is confined to Approach 1: the Integrated Gradients token selection is performed on the validation split, and Table 2 reports the fairness metrics on that same split. The token mask is thus fitted to the evaluation set, making the reported D_KL reduction an in-sample statistic rather than an unbiased estimate on unseen resumes; the paper reports no separate held-out test set. Because the paper's strongest conclusion ('both methods successfully reduced gender information') depends on Approach 1's Table 2, that specific component is partly a fitted result. Approach 2 is not affected by this leak and shows a real, smaller reduction (ratio about 0.84, recall gap about 6.8 points), so the overall claim retains independent content; score 6 reflects this partial circularity. The manuscript also has empty section cross-references ('Sect.' and 'see Section'), which are production issues and do not affect the circularity assessment.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a synthetic dataset with artificially injected bias, hand-chosen hyperparameters (especially lambda, token counts, top-500), and the assumption that removing gender information from inputs or latent spaces does not destroy task-relevant signal. No new entities are introduced.

free parameters (5)
  • lambda (adversarial trade-off) = 0.1
    Balance between RMSE loss and mutual information term in Eq. 1; chosen by hand with no sensitivity analysis, directly affects the debiasing strength in Approach 2.
  • Top-500 shortlist threshold = 500
    All fairness metrics (proportions, D_KL, recall) are computed on the top-500 candidates; the results are sensitive to this arbitrary cutoff, which is not varied.
  • Number of relevant tokens per resume and per group = 20 and 30
    Approach 1 extracts 20 most relevant tokens per resume and 30 per gender/sector group; these hand-chosen values determine which words are masked and hence the observed mitigation.
  • Training hyperparameters = 10 epochs, batch 32, lr 1e-3, dropout 0.3, hidden units 300/20
    Standard but manually selected with no search; all reported results depend on these choices.
  • Gender bias strength in y_G = not reported in this paper
    The magnitude of the injected bias in FairCVdb's gender-biased scores is inherited from prior work (Pena et al. 2020a,b) and is not quantified here, so the stress level of the test is undefined.
assumptions (4)
  • domain assumption FairCVdb synthetic resumes and artificial biased scores faithfully represent real recruitment data and annotation bias
    The entire experimental setup uses this synthetic testbed (Materials and Methods, Dataset: FairCVdb); no real-world validation is provided, so results may not transfer.
  • domain assumption Gender is binary and correctly labeled in the data
    Footnote 3 acknowledges this limitation, but all experiments and metrics rely on binary z.
  • domain assumption Removing gender information does not remove task-relevant information sufficient for accurate scoring
    The mitigation is evaluated on recall; if masked tokens carry job-relevant content, utility suffers. The paper notes recall for Approach 1 falls below the unbiased baseline, indicating partial violation.
  • domain assumption Integrated Gradients attributions reliably identify gender-proxy tokens
    Approach 1 assumes the token relevance scores expose the actual bias carriers; this is not quantitatively validated against ground truth bias tokens.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Addressing Bias in LLMs: Strategies and Application to Fair AI-based Recruitment." pith.science (2026). https://pith.science/paper/C2ZHJHWI

@misc{pith2026250611880,
  author       = {Pith},
  title        = {Pith review of: Addressing Bias in LLMs: Strategies and Application to Fair AI-based Recruitment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C2ZHJHWI}},
  note         = {Machine review of arXiv:2506.11880}
}
read the original abstract

The use of language technologies in high-stake settings is increasing in recent years, mostly motivated by the success of Large Language Models (LLMs). However, despite the great performance of LLMs, they are are susceptible to ethical concerns, such as demographic biases, accountability, or privacy. This work seeks to analyze the capacity of Transformers-based systems to learn demographic biases present in the data, using a case study on AI-based automated recruitment. We propose a privacy-enhancing framework to reduce gender information from the learning pipeline as a way to mitigate biased behaviors in the final tools. Our experiments analyze the influence of data biases on systems built on two different LLMs, and how the proposed framework effectively prevents trained systems from reproducing the bias in the data.

Figures

Figures reproduced from arXiv: 2506.11880 by the authors.

Figure 1
Figure 1. Learning architecture for the resume scoring tool [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Proposed privacy-enhancing learning framework for bias mitigation. We aim to mitigate the effect of demographic [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Visualization of the t-SNE projections for the [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Privacy-Aware Detection of Fake Identity Documents: Methodology, Benchmark, and Improved Algorithms (FakeIDet2)

    cs.CR 2025-08 conditional novelty 6.0 of 10

    Patch-based training on anonymized ID fragments detects physical and synthetic fake IDs across unseen datasets without sharing full identity documents.

  2. PBa-LLM: Privacy- and Bias-aware NLP using Named-Entity Recognition (NER)

    cs.CL 2025-06 conditional novelty 4.0 of 10

    NER-based removal of person and location entities from resumes preserves occupancy-prediction accuracy on FairCVdb, and combined with a debiasing module yields gender-balanced shortlists.

Reference graph

Works this paper leans on

61 extracted references · 56 canonical work pages · cited by 2 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    G.; and Blank, R

    Altonji, J. G.; and Blank, R. 1999. Race and Gender in the Labor Market. In Handbook of Labor Economics, volume 3, chapter 10, 3143--3259. Elsevier

    work page 1999

  4. [4]

    W.; and Romero, J

    Bendick Jr, M.; Jackson, C. W.; and Romero, J. H. 1997. Employment Discrimination Against Older Workers: An Experimental Study of Hiring Practices . Journal of Aging & Social Policy, 8: 25--46

    work page 1997

  5. [5]

    Bertrand, M.; and Mullainathan, S. 2004. Are Emily and Greg more Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination . American Economic Review, 94(4): 991--1013

    work page 2004

  6. [6]

    Blodgett, S.; Barocas, S.; Daum \'e III, H.; and Wallach, H. 2020. Language (Technology) is Power: A Critical Survey of ``Bias'' in NLP . In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), 5454--5476

    work page 2020

  7. [7]

    Bogen, M.; and Rieke, A. 2018. Help Wanted: Examination of Hiring Algorithms, Equity, and Bias . Technical report

    work page 2018

  8. [8]

    Bolukbasi, T.; Chang, K.; Zou, J.; Saligrama, V.; et al. 2016. Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings . In Proceedings of the International Conference on Neural Information Processing Systems (NIPS 2016), 4356--4364

    work page 2016

Show all 61 references
  1. [9]

    Bommasani, R.; Klyman, K.; Zhang, D.; and Liang, P. 2023. Do Foundation Model Providers Comply with the EU AI Act?

    2023

  2. [10]

    Caliskan, A.; Bryson, J.; and Narayanan, A. 2017. Semantics Derived Automatically from Language Corpora Contain Human-Like Biases . Science, 356(6334): 183--186

    2017

  3. [11]

    Cao, Y.; and Daum \'e , H. 2021. Toward Gender-Inclusive Coreference Resolution: An Analysis of Gender and Bias Throughout the Machine Learning Lifecycle . Computational Linguistics, 47(3): 615--661

    2021

  4. [12]

    Chen, Y.; Mahoney, C.; Grasso, I.; Wali, E.; et al. 2021. Gender Bias and Under-Representation in Natural Language Processing Across Human Languages . In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 24--34

    2021

  5. [13]

    R.; and Liu, H

    Cheng, L.; Varshney, K. R.; and Liu, H. 2021. Socially Responsible AI Algorithms: Issues, Purposes, and Challenges . Journal of Artificial Intelligence Research, 71: 1137--1181

    2021

  6. [14]

    Commission, U. E. E. O.; et al. 1979. Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures . Federal Register, 40(43)

    1979

  7. [15]

    Dastin, J. 2018. Amazon Scraps Secret AI Recruiting Tool that Showed Bias Against Women . Reuters

    2018

  8. [16]

    De-Arteaga, M.; Romanov, R.; Wallach, H.; Chayes, J.; et al. 2019. Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting . In Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), 120--128

    2019

  9. [17]

    DeAlcala, D.; Serna, I.; Morales, A.; Fierrez, J.; et al. 2023. Measuring Bias in AI Models: An Statistical Approach Introducing N-Sigma . In 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC), 1167--1172

    2023

  10. [18]

    Devlin, J.; Chang, M.; Lee, K.; and Toutanova, K. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding . In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologi...

    2019

  11. [19]

    Dixon, L.; Li, J.; Sorensen, J.; Thain, N.; et al. 2018. Measuring and Mitigating Unintended Bias in Text Classification . In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 67–73

    2018

  12. [20]

    Executive Office of the President . 2016. Big Data: A report on Algorithmic Systems, Opportunity, and Civil Rights. Technical report, The White House

    2016

  13. [21]

    Garg, N.; Schiebinger, L.; Jurafsky, D.; and Zou, J. 2018. Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypes . Proceedings of the National Academy of Sciences, 115(16): E3635--E3644

    2018

  14. [22]

    Ge, S.; Zhou, C.; Hou, R.; Khabsa, M.; et al. 2023. MART: Improving LLM Safety with Multi-round Automatic Red-Teaming . Preprint at https://arxiv.org/abs/2311.07689

    2023 arXiv

  15. [23]

    Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; et al. 2014. Generative Adversarial Nets . In Proceedings of the International Conference on Neural Information Processing Systems (NIPS 2014), volume 27, 2672--2680

    2014

  16. [24]

    Hardt, M.; Price, E.; and Srebro, N. 2016. Equality of Opportunity in Supervised Learning. In Proceedings of the International Conference on Neural Information Processing Systems (NIPS), 3323--3331

    2016

  17. [25]

    Hassanpour, A.; Moradikia, M.; Yang, B.; Abdelhadi, A.; et al. 2022. Differential Privacy Preservation in Robust Continual Learning. IEEE Access, 10: 24273--2428

    2022

  18. [26]

    Hu, E.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; et al. 2021. LoRA: Low-Rank Adaptation of Large Language Models . In Proceeding of the International Conference on Learning Representations (ICLR)

    2021

  19. [27]

    Keyes, O. 2018. The Misgendering Machines: Trans/HCI Implications of Automatic Gender Recognition . Proceedings of ACM Human-Computer Interaction, 2(CSCW): 1--22

    2018

  20. [28]

    Kim, B.; Kim, H.; Kim, K.; Kim, S.; et al. 2019. Learning Not to Learn: Training Deep Neural Networks with Biased Data . In Proceedings of the IEEE/CVF International Conference on Computer Vision and Pattern Recognition (CVPR), 9012--9020

    2019

  21. [29]

    Kiritchenko, S.; and Mohammad, S. 2018. Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems . In Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics, 43--53. New Orleans, Louisiana: Association for Computational Linguistics

    2018

  22. [30]

    Leavy, S.; Siapera, E.; and O'Sullivan, B. 2021. Ethical Data Curation for AI: An Approach Based on Feminist Epistemology and Critical Theories of Race . In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (AIES), 695--703

    2021

  23. [31]

    Liu, Y.; Ott, M.; Goyal, N.; Du, J.; et al. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach . arXiv/1907.11692

    2019 arXiv

  24. [32]

    R.; and Rudinger, R

    May, C.; Wang, A.; Bordia, S.; Bowman, S. R.; and Rudinger, R. 2019. On Measuring Social Biases in Sentence Encoders . In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), 622--628

    2019

  25. [33]

    Morales, A.; Fierrez, J.; Vera-Rodriguez, R.; and Tolosana, R. 2021. SensitiveNets: Learning Agnostic Representations with Application to Face Recognition . IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(6): 2158--2164

    2021

  26. [34]

    OpenAI. 2023. GPT-4 Technical Report . arXiv/2303.08774

    2023 arXiv

  27. [35]

    Park, J.; Shin, J.; and Fung, P. 2018. Reducing Gender Bias in Abusive Language Detection . In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2799--2804

    2018

  28. [36]

    Pe \ n a, A.; Morales, A.; Fierrez, J.; Serna, I.; et al. 2023. Leveraging Large Language Models for Topic Classification in the Domain of Public Affairs . In Document Analysis and Recognition -- ICDAR 2023 Workshops, 20 -- 33

    2023

  29. [37]

    Peña, A.; Fierrez, J.; Lapedriza, A.; and Morales, A. 2021. Learning Emotional-Blinded Face Representations . In IAPR Intl. Conf. on Pattern Recognition (ICPR), 3566--3573

    2021

  30. [38]

    Peña, A.; Serna, I.; Morales, A.; and Fierrez, J. 2020 a . Bias in Multimodal AI: Testbed for Fair Automatic Recruitment . In IEEE/CVF Conf. on Computer Vision and Pattern Recognition Workshops (CVPRw). Also presented at ICML 2020 Workshop on Human-in-the-Loop Learning

    2020

  31. [39]

    Peña, A.; Serna, I.; Morales, A.; and Fierrez, J. 2020 b . FairCVtest Demo: Understanding Bias in Multimodal Learning with a Testbed in Fair Automatic Recruitment . In ACM Intl. Conf. on Multimodal Interaction (ICMI), 760--761. Also presented at European Conf. on Computer Visi...

    2020

  32. [40]

    Peña, A.; Serna, I.; Morales, A.; Fierrez, J.; et al. 2023. Human-Centric Multimodal Machine Learning: Recent Advances and Testbed on AI -based Recruitment . SN Computer Science, 4(43)

    2023

  33. [41]

    Prates, M.; Avelar, P.; and Lamb, L. 2020. Assessing Gender Bias in Machine Translation: A Case Study with Google Translate . Neural Computing and Applications, 32: 6363--6381

    2020

  34. [42]

    M.; et al

    Qian, R.; Ross, C.; Fernandes, J.; Smith, E. M.; et al. 2022. Perturbation Augmentation for Fairer NLP . In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 9496--9521

    2022

  35. [43]

    Romanov, A.; De-Arteaga, M.; Wallach, H.; Chayes, J.; et al. 2019. What's in a Name? Reducing Bias in Bios without Access to Protected Attributes . In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language T...

    2019

  36. [44]

    Rudinger, R.; May, C.; and Van Durme, B. 2017. Social Bias in Elicited Natural Language Inferences . In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, 74--79

    2017

  37. [45]

    S \'a nchez, J.; Dencik, L.; and Edwards, L. 2020. What Does It Mean to 'Solve' the Problem of Discrimination in Hiring? Social, Technical and Legal Perspectives from the UK on Automated Hiring Systems . In Proceedings of the ACM Conference on Fairness, Accountability, and Tra...

    2020

  38. [46]

    Sap, M.; Card, D.; Gabriel, S.; Choi, Y.; et al. 2019. The Risk of Racial Bias in Hate Speech Detection . In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), 1668--1678

    2019

  39. [47]

    S.; Mattei, N.; and Dickerson, J

    Schumann, C.; Foster, J. S.; Mattei, N.; and Dickerson, J. P. 2020. We Need Fairness and Explainability in Algorithmic Hiring . In Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems, 1716--1720

    2020

  40. [48]

    Serna, I.; DeAlcala, D.; Morales, A.; Fierrez, J.; and Ortega-Garcia, J. 2022 a . IFBiD: Inference-Free Bias Detection . In AAAI Workshop on Artificial Intelligence Safety (SafeAI), volume 3087 of CEUR

    2022

  41. [49]

    Serna, I.; Morales, A.; Fierrez, J.; Cebrian, M.; Obradovich, N.; and Rahwan, I. 2020. Algorithmic Discrimination: Formulation and Exploration in Deep Learning-based Face Biometrics . In AAAI Workshop on Artificial Intelligence Safety (SafeAI), volume 2560 of CEUR-WS, 146--152

    2020

  42. [50]

    Serna, I.; Morales, A.; Fierrez, J.; and Obradovich, N. 2022 b . Sensitive Loss: Improving Accuracy and Fairness of Face Representations with Discrimination-Aware Deep Learning . Artificial Intelligence, 305: 103682

    2022

  43. [51]

    Serna, I.; Peña, A.; Morales, A.; and Fierrez, J. 2021. InsideBias: Measuring Bias in Deep Networks and Application to Face Gender Biometrics . In Proceedings of IAPR International Conference on Pattern Recognition (ICPR), 3720--3727. Milan, Italy

    2021

  44. [52]

    Sheng, E.; Chang, K.; Natarajan, P.; and Peng, N. 2019. The Woman Worked as a Babysitter: On Biases in Language Generation . In Proceedings of the Conference on Empirical Methods in Natural Language Processing and the International Joint Conference on Natural Language Processi...

    2019

  45. [53]

    Stanovsky, G.; Smith, N.; and Zettlemoyer, L. 2019. Evaluating Gender Bias in Machine Translation . In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), 1679--1684

    2019

  46. [54]

    Sundararajan, M.; Taly, A.; and Yan, Q. 2017. Axiomatic Attribution for Deep Networks. In Proceedings of the International Conference on Machine Learning (ICML), 3319--3328. PMLR

    2017

  47. [55]

    Touvron, H.; Martin, L.; Stone, K.; Albert, P.; et al. 2023. LLaMA 2: Open Foundation and Fine-Tuned Chat Models . Technical report, Meta AI

    2023

  48. [56]

    Van der Maaten, L.; and Hinton, G. 2008. Visualizing Data Using t-SNE . Journal of Machine Learning Research, 9(86): 2579--2605

    2008

  49. [57]

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; et al. 2017. Attention is All You Need . In Proceedings of the International Conference on Neural Information Processing Systems (NIPS), volume 30, 6000--6010

    2017

  50. [58]

    A.; McKenna, M.; and Chen, J

    Watkins, E. A.; McKenna, M.; and Chen, J. 2022. The Four-Fifths Rule is not Disparate Impact: A Woeful Tale of Epistemic Trespassing in Algorithmic Fairness. arXiv/2202.09519

    2022 arXiv

  51. [59]

    Yang, K.; and Stoyanovich, J. 2017. Measuring Fairness in Ranked Outputs . In Proceedings of the International Conference on Scientific and Statistical Database Management, 1--6

    2017

  52. [60]

    Zehlike, M.; Bonchi, F.; Castillo, C.; Hajian, S.; Megahed, M.; and Baeza-Yates, R. 2017. FA*IR: A Fair Top-K Ranking Algorithm . In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, 1569--1578

    2017

  53. [61]

    Zhao, J.; Wang, T.; Yatskar, M.; Ordonez, V.; and Chang, K. 2018. Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods . In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technolog...

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.