Pith. sign in

REVIEW 4 major objections 6 minor 3 cited by

Mitigate One, Skew Another? Tackling Intersectional Biases in Text-to-Image Models

T0 review · 4 major / 6 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 BiasConnect estimates how mitigating bias on one axis (say gender) will change the distribution of another axis (say age) before any mitigation is run, and InterMit uses those estimates to guide faster, higher-quality multi-axis debiasing.

desk verdict Useful tool with a real validation gap: the +0.65 correlation is inflated by a shared pre-mitigation baseline, so the headline claim needs a re-analysis. read the letter →

arxiv 2505.17280 v1 pith:7MU2IMAU submitted 2025-05-22 cs.CV

Pushkar Shukla , Aditya Chinchure , Emily Diana , Alexander Tolbert , Kartik Hosanagar , Vineeth N Balasubramanian , Leonid Sigal , Matthew Turk This is my paper
classification cs.CV
keywords intersectionalbiastext-to-imagegenerationmitigationcounterfactualpromptsSensitivityWassersteindistancefairnessauditing
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 claims that biases in text-to-image models are not independent: intervening on one bias axis, such as gender, can improve or worsen another axis, such as age or ethnicity. To make this measurable, it introduces Intersectional Sensitivity (IS), a score computed from counterfactual prompts that estimates whether mitigating bias along axis $B_x$ will move axis $B_y$ closer to or further from an ideal distribution. The paper reports that IS predictions correlate with observed post-mitigation outcomes at +0.65, and that its own mitigation algorithm, InterMit, reaches lower overall bias with fewer steps and better image quality than a standard baseline. If these estimates hold across models, bias mitigation can be planned and audited before expensive model changes.

What carries the argument

Intersectional Sensitivity ($IS_{xy}$), computed by replacing the initial distribution of attributes on axis $B_y$ with the sum of those distributions across all counterfactual prompts of $B_x$, each weighted equally. This reweighting simulates what a mitigation method would do by enforcing equal representation along $B_x$; the difference in Wasserstein distance to the ideal distribution before and after the reweighting is the sensitivity score. The paper also assembles these scores into a Bias Intersectionality Matrix $S$, whose rows and columns are bias axes, and InterMit iteratively picks the axis $i^* = \arg\max_i \langle s'_i, p \rangle$ against the user's priority vector $p$ to decide which bias to mitigate next.

What would settle it

Run BiasConnect on a text-to-image model not used in the paper, such as an open SDXL or a closed API model, compute predicted IS for all axis pairs on the 26 occupation prompts, then actually mitigate each axis with a training-based method and compute observed IS. If the average correlation between predicted and observed IS falls below roughly 0.5, the equal-proportion reweighting proxy does not transfer.

Watch

Extended reading notes

Core claim

BiasConnect's central claim is that the effect of mitigating one bias axis on another can be estimated without performing any mitigation, by reweighting the target axis's attribute distribution across equal-proportion counterfactuals of the intervention axis. Formally, if $B_x$ is the axis being mitigated and $B_y$ the axis being observed, the paper defines $IS_{xy} = w^{init}_{B_y} - w^{B_x}_{B_y}$, where each $w$ is the Wasserstein distance from the empirical attribute distribution to a user-chosen ideal distribution. A positive value means mitigating $B_x$ improves diversity on $B_y$; a negative value means it worsens it. The paper validates this estimate by mitigating all axes with ITI-GEN on Stable Diffusion 1.4 and finds an average correlation of +0.65 between predicted and observed IS, with particularly high values for musician, accountant, and lawyer. On top of this, InterMit uses the resulting interaction matrix to select, at each step, the axis whose mitigation is most aligned with the user's priority vector, and it reports lower bias (0.33 vs. 0.52), fewer steps (2.38 vs. 3.15), and higher image quality than ITI-GEN.

Load-bearing premise

The equal-proportion reweighting of counterfactual prompts along $B_x$ is assumed to faithfully mimic what an actual mitigation method does to the distribution of $B_y$.

Editorial extensions

If this is right

  • Model auditors can use BiasConnect to identify, for a given occupation prompt, which bias axes are coupled and in which direction, without generating additional images.
  • InterMit's step selection can reduce the total number of mitigation runs: in the paper's examples mitigating ethnicity alone also reduces age bias due to demographic overlap, cutting the number of steps.
  • The framework is training-free and modular, so it can be layered on top of existing sequential debiasing methods, not just prompt modification.
  • When the ideal distribution is set to a real-world reference instead of uniform, sensitivity scores can flip sign, showing that conclusions about intersectional impact depend on the chosen target distribution.

Reading between the lines

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

  • The +0.95 correlation reported for prompt-modification mitigation is expected, because prompt modification and the sensitivity estimate both use the same counterfactual reweighting; a stronger validation would apply BiasConnect to a training-based mitigation on a model other than Stable Diffusion 1.4.
  • If the equal-proportion reweighting proxy fails for non-uniform target distributions, the IS estimate would need to be re-derived for each new target distribution; the paper only tests uniform and a single real-world ideal.
  • The approach treats 'mitigation' as equal representation across predefined attribute levels; it does not address axes where the relevant attribute set is open-ended, though the paper tests dynamic axes from the TIBET dataset with LLM-generated questions.
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 / 6 minor

Summary. The paper introduces BiasConnect, a counterfactual-prompt analysis framework for text-to-image (TTI) models, together with a metric called Intersectional Sensitivity (IS) that aims to quantify how intervening on one bias axis (Bx) changes the distribution of another bias axis (By). The authors claim that predicted IS values correlate strongly (+0.65) with observed post-mitigation outcomes, and they present this as evidence that BiasConnect can estimate the effect of mitigation before performing expensive model changes. Building on IS, the paper proposes InterMit, an iterative mitigation algorithm that uses the Bias Intersectionality Matrix and user-specified priority vectors to sequentially mitigate multiple bias axes. InterMit is evaluated against ITI-GEN on Stable Diffusion 1.4 and on Stable Diffusion 3.5, reporting lower residual bias, fewer mitigation steps, and higher image quality.

Significance. If the central prediction claim were valid, the paper would make a practically useful contribution: it would allow researchers and practitioners to audit and plan bias mitigation in TTI models without running full mitigation pipelines, and it would provide a modular framework for intersectional mitigation. The proposed IS metric is simple and interpretable, and the paper includes counterfactual analyses across several models, including Stable Diffusion 1.4, SD 3.5, Flux, Playground, and Kandinsky, as well as a robustness study. The InterMit algorithm is training-free and modular, which are genuine strengths. However, the validity of the headline +0.65 correlation is the load-bearing point, and the current evidence is weaker than the paper claims.

major comments (4)
  1. [Sec. 5.3, Eqs. (4), (6), (8)] The reported validation correlation of +0.65 is computed between IS_xy = w_init_By - w_Bx_By and IS_mit(x)_xy = w_init_By - w_mit(x)_By. Both quantities contain the same initial Wasserstein distance w_init_By, so the reported correlation can be large even if the predicted post-intervention distribution w_Bx_By has no ability to predict the actual post-mitigation distribution w_mit(x)_By. This is the classic shared-baseline artifact: corr(A-X, A-Y) is inflated by variance in A. The paper does not report corr(w_Bx_By, w_mit(x)_By) or a partial correlation controlling for w_init_By, so the abstract's central claim is not actually supported by the reported statistic. Please report the baseline-controlled correlations, along with scatterplots and confidence intervals.
  2. [Sec. 5.3 and Appendix A.7] The +0.95 correlation for PM is near-tautological. The PM mitigation method described in Appendix A.7 mitigates an axis by generating equal proportions of counterfactual prompts for that axis, which is exactly the operation used to construct D_Bx_By in Sec. 3.3. The paper acknowledges this only as 'unsurprising, as it uses similar counterfactual prompts for mitigation' (Sec. 5.3), but the relationship is stronger than similarity: the predicted and 'observed' quantities are built from the same reweighting operation. This validation cannot independently confirm that the counterfactual proxy predicts a real mitigation method.
  3. [Sec. 5.3 and Abstract] After removing the PM validation as essentially definitional, the only non-circular evidence for the central claim is the ITI-GEN experiment on Stable Diffusion 1.4, and that single result is confounded by the shared baseline w_init_By. The paper does not report the unconfounded correlation for the ITI-GEN experiment, and it does not validate the IS prediction on any other model with an independent mitigation method. Since the claim that 'these estimates show strong correlation (+0.65) with observed post-mitigation outcomes' is the paper's main advertised result, this gap must be addressed with a baseline-controlled statistic and, ideally, additional independent mitigation evaluations on more than one model.
  4. [Sec. 5.5 and Table 1] The InterMit evaluation compares aggregate mitigation amount and steps against ITI-GEN, but the comparison is not apples-to-apples: ITI-GEN is forced to mitigate all biases in the priority vector, whereas InterMit can stop early using the threshold epsilon. The paper does not report variance or significance across repeated runs, and the SD3.5 results use a different prompt set and priority vector, as the table footnote states. This makes it difficult to judge whether the reported improvements in bias and steps are robust or partly an artifact of the evaluation protocol.
minor comments (6)
  1. [Sec. 3.3] The statement 'w_By is computed by normalizing w_By based on the number of counterfactuals in By' is unclear: it is not obvious how the Wasserstein distance is normalized to lie in [0,1], and the dependence on the number of counterfactuals is not formalized.
  2. [Sec. 4 and Algorithm 1] The bias threshold epsilon is set empirically to 0.35 for all experiments, but no sensitivity analysis is provided. Since epsilon controls the stopping criterion, it can directly affect the reported MitSteps and MitAmt values.
  3. [Fig. 3 caption] The caption contains a duplicated phrase: 'BiasConnect can allow the user the user to understand' should read 'allow the user to understand'.
  4. [Sec. 5.4] The sentence 'an removing 32 images (66.6 %) yields a 31.3%' should be 'removing 32 images'.
  5. [Sec. 5.2] The prompt-level examples are useful, but the IS values cited in the text (e.g., IS = 0.31, IS = -0.40) are presented without error bars or a sensitivity analysis, so it is hard to tell which differences are meaningful given the VQA error rates reported in Sec. 5.4.
  6. [Appendix A.9] The appendix repeatedly uses the phrase 'this new score can be defined as' and then gives equations identical to those in Sec. 3.3 and A.5. The redundancy is fine, but it would help to explicitly state which quantities are predicted and which are observed, and to label PM as a same-reweighting sanity check.

Circularity Check

2 steps flagged · score 4.0 of 10

The +0.95 PM validation is the same reweighting operation as the prediction, and the +0.65 ITI-GEN validation shares the winit_By baseline in both correlated quantities; the counterfactual proxy itself remains an independent, testable construction.

  1. self definitional [Sec. 5.3; Appendix A.7 (PM definition)]
    ""An average correlation of +0.95 using PM is unsurprising, as it uses similar counterfactual prompts for mitigation." (Sec. 5.3). Appendix A.7: "All future mitigation steps will add to these permutations, and an equal number of images are generated for each prompt in the set, such that the total is 48 (or more) images.""

    The predicted quantity w_Bx_By is computed in Sec. 3.3 by summing the By distributions across all counterfactuals of Bx, i.e., equal-proportion reweighting over Bx. The PM mitigation method in Appendix A.7 generates an equal number of images for every permutation of counterfactual prompts of the mitigated axis. When PM mitigates Bx alone, the resulting By distribution is, up to finite-sample noise, exactly the same D_Bx_By used to define the predicted IS in Eq. (4). Hence IS_xy and the "observed" post-PM IS_mit(x)_xy coincide by construction, so a +0.95 correlation is a consistency check of the implementation rather than independent validation. The paper's own "unsurprising" remark acknowledges this.

  2. other [Appendix A.9, Sec. 5.3, Eqs. (4), (7), (8)]
    ""To quantify the effectiveness of BiasConnect we measure the average correlation between the Intersectional Sensitivity scores before IS xy and after mitigation IS mit(x) xy across all intersectional relationships Bx → By present for each prompt." The scores are defined as IS xy = winit By − wBx By (4) and, with wBx By = W1(Dmit(Bx) By, D*) (7), IS mit(x) xy = winit By − wBx By (8)."

    Both correlated quantities subtract the same measured initial distance winit_By: predicted IS_xy = winit_By − w_Bx_By and post-mitigation IS_mit(x)_xy = winit_By − w_mit(x)_By. Since winit_By varies across the 26 occupations and 8 axes, the shared term inflates corr(IS_xy, IS_mit(x)_xy) even if w_Bx_By and w_mit(x)_By are unrelated. The paper does not report corr(w_Bx_By, w_mit(x)_By) or a partial correlation controlling for winit_By, so the abstract's +0.65 claim is not established as a test of the predictive component. The reported statistic partially reduces to the shared initial-bias baseline rather than to the counterfactual proxy's predictive content.

full rationale

BiasConnect's IS estimator is not fitted to post-mitigation data: it is a counterfactual reweighting proxy (equal-proportion mixing over Bx counterfactuals) and is therefore an independently testable construction. The strongest independent evidence is the ITI-GEN comparison, since ITI-GEN is a learned mitigation method distinct from the equal-proportion reweighting used in the estimator; however, the reported +0.65 correlation is confounded because both the predicted and observed IS values subtract the same winit_By term, and the paper does not provide a baseline-controlled correlation. The second validation, +0.95 with PM, is explicitly acknowledged in the manuscript as unsurprising and is near-tautological: PM is exactly the equal-proportion reweighting operation used to define the prediction, so this check does not add independent confirmation. The paper's own admission lowers the severity because the authors do not present the PM result as independent. InterMit's selection rule uses the same matrix that PM's mitigation realizes, so the mitigation experiments are self-consistent but do not independently validate the proxy; they are not a load-bearing circular derivation. TIBET (Chinchure et al., 2024) is prior work by overlapping authors, but it is used as a dataset and VQA tooling and is not the justification for IS or for the mitigation claims, so no self-citation circularity is present. Overall, there is partial circularity in the validation and a shared-baseline artifact in the headline statistic, but the core estimator has independent content, warranting a moderate score of 4 rather than 6 or higher.

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

The central machinery rests on hand-set parameters (epsilon, image count, uniform ideal distribution) and on the author-defined counterfactual template sets. No new physical entities are introduced. The most consequential axiom is that equal-proportion reweighting of counterfactuals faithfully simulates real mitigation methods.

free parameters (4)
  • bias threshold epsilon = 0.35
    Set empirically in Section 5.1; determines when InterMit stops, so it directly controls the reported MitSteps and final MitAmt (0.33 vs 0.52).
  • images per prompt = 48
    Chosen for all generation experiments (Section 5.1); robustness test shows IS changes by 10.5% when 8 images are removed.
  • ideal distribution D* = uniform over attribute choices
    Modeling choice for all main experiments (Section 3.3); the paper itself shows flipping to a real-world distribution changes IS signs.
  • counterfactual sets per axis = 2-6 attribute values
    Author-defined templates in Appendix A.1; axes are reduced to a small fixed set of attributes, which defines what counts as bias.
assumptions (5)
  • domain assumption VQA outputs (MiniGPT-v2) are treated as ground-truth attributes for bias axes.
    Section 3.2; robustness analysis simulates random VQA errors at 5-40% rather than measuring the actual error rate on this data, and notes TIBET observed 18% error.
  • domain assumption Equal-proportion reweighting across counterfactuals of Bx faithfully simulates the effect of a real mitigation method on By.
    Section 3.3: intervention ensures all counterfactuals of Bx are equally represented; this is exactly the PM procedure, making the +0.95 PM correlation near-tautological. For ITI-GEN it is an untested approximation, validated only on SD1.4.
  • ad hoc to paper The bias axes are independent within a query and each is captured by a fixed author-chosen attribute set.
    The 8 occupation axes and their choices (e.g., gender=male/female) are defined in Appendix A.1, excluding many other identities and attribute values.
  • domain assumption Dot-product alignment of the sensitivity matrix row with the priority vector is a reliable way to select the next mitigation axis.
    Algorithm 1 line 8; no theoretical or empirical comparison to other selection rules is provided.
  • domain assumption Wasserstein-1 distance to the ideal distribution is a valid measure of bias along an axis.
    Section 3.3, Eqs. 1-4; sensible but one of many possible distributional distance choices, and the normalization to [0,1] based on the number of counterfactuals is heuristic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mitigate One, Skew Another? Tackling Intersectional Biases in Text-to-Image Models." pith.science (2026). https://pith.science/paper/7MU2IMAU

@misc{pith2026250517280,
  author       = {Pith},
  title        = {Pith review of: Mitigate One, Skew Another? Tackling Intersectional Biases in Text-to-Image Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7MU2IMAU}},
  note         = {Machine review of arXiv:2505.17280}
}
read the original abstract

The biases exhibited by text-to-image (TTI) models are often treated as independent, though in reality, they may be deeply interrelated. Addressing bias along one dimension - such as ethnicity or age - can inadvertently affect another, like gender, either mitigating or exacerbating existing disparities. Understanding these interdependencies is crucial for designing fairer generative models, yet measuring such effects quantitatively remains a challenge. To address this, we introduce BiasConnect, a novel tool for analyzing and quantifying bias interactions in TTI models. BiasConnect uses counterfactual interventions along different bias axes to reveal the underlying structure of these interactions and estimates the effect of mitigating one bias axis on another. These estimates show strong correlation (+0.65) with observed post-mitigation outcomes. Building on BiasConnect, we propose InterMit, an intersectional bias mitigation algorithm guided by user-defined target distributions and priority weights. InterMit achieves lower bias (0.33 vs. 0.52) with fewer mitigation steps (2.38 vs. 3.15 average steps), and yields superior image quality compared to traditional techniques. Although our implementation is training-free, InterMit is modular and can be integrated with many existing debiasing approaches for TTI models, making it a flexible and extensible solution.

Figures

Figures reproduced from arXiv: 2505.17280 by the authors.

Figure 1
Figure 1. An example for which BiasConnect estimates a negative impact of bias mitigation along one axis on another axis. For this query, increasing the gender diver￾sity (Gen) skews age distribution (Age) for images of musicians generated by Flux-dev. non-social dimensions – including gender, race, clothing, and age – leading to skewed or inaccu￾rate representations. As a result, TTI models may reinforce harmful stereotypes … view at source ↗
Figure 2
Figure 2. An overview of BiasConnect. We use a counterfactual-based approach to measure how interventions along a single bias axes impact other bias axes. Our metric Intersectional Sensitivity estimates how bias mitigation on one axis impacts another. Our results are visualized as a a matrix called the Bias Intersectionality Matrix. and datasets, addressing biases in various language￾vision tasks. These investigations have ex… view at source ↗
Figure 3
Figure 3. Analyzing bias intersectionality matrices from [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Three examples of mitigation using InterMit. Priority vectors guide the mitigation process. Columns that are a part of sub-matrix S ′ are in blue. As shown in (a) and (c), the algorithm mitigates multiple biases in fewer steps. (b) shows how user-defined priorities gui…

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Understanding and evaluating computer vision models through the lens of counterfactuals

    cs.CV 2025-08 conditional novelty 6.0 of 10

    Counterfactual-based methods for concept attribution in classifiers and for dynamic bias evaluation and mitigation in text-to-image models.

  2. Adultification Bias in LLMs and Text-to-Image Models

    cs.CY 2025-06 conditional novelty 6.0 of 10

    Large language and text-to-image models show measurable adultification bias, portraying Black girls as more mature, culpable, and sexualized than White girls in several tested models.

  3. LLM Harms: A Taxonomy and Discussion

    cs.CY 2025-12 unverdicted novelty 3.0 of 10

    This paper proposes a taxonomy of LLM harms in five categories and suggests mitigation strategies plus a dynamic auditing system for responsible development.

Reference graph

Works this paper leans on

63 extracted references · 44 canonical work pages · cited by 3 Pith papers

  1. [1]

    Lavisha Aggarwal and Shruti Bhargava. 2023. Fairness in ai systems: Mitigating gender bias from language-vision models. arXiv preprint arXiv:2305.01888

    arXiv 2023

  2. [2]

    Jaimeen Ahn and Alice Oh. 2021. Mitigating language-dependent ethnic bias in bert. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 533--549

    work page 2021

  3. [3]

    Xuechunzi Bai, Angelina Wang, Ilia Sucholutsky, and Thomas L Griffiths. 2025. Explicitly unbiased large language models still form biased associations. Proceedings of the National Academy of Sciences, 122(8):e2416228122

    2025

  4. [4]

    Emily M Bender, Timnit Gebru, Angelina McMillan-Major, and Margaret Shmitchell. 2021. On the dangers of stochastic parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pages 610--623

    work page 2021

  5. [5]

    Federico Bianchi, Pratyusha Kalluri, Esin Durmus, Faisal Ladhak, Myra Cheng, Debora Nozza, Tatsunori Hashimoto, Dan Jurafsky, James Zou, and Aylin Caliskan. 2023. Easily accessible text-to-image generation amplifies demographic stereotypes at large scale. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, pages 1493--1504

    work page 2023

  6. [6]

    Abeba Birhane and Vinay Uday Prabhu. 2021. Multimodal datasets: Misogyny, pornography, and malignant stereotypes. arXiv preprint arXiv:2110.01963

    arXiv 2021

  7. [7]

    BlackForestLabs. 2024. Flux. https://github.com/black-forest-labs/flux

    work page 2024

  8. [8]

    Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. Advances in neural information processing systems, 29

    2016

Show all 63 references
  1. [9]

    Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency, pages 77--91. PMLR

    2018

  2. [10]

    Jun Chen, Deyao Zhu, Xiaoqian Shen, Xiang Li, Zechun Liu, Pengchuan Zhang, Raghuraman Krishnamoorthi, Vikas Chandra, Yunyang Xiong, and Mohamed Elhoseiny. 2023. Minigpt-v2: large language model as a unified interface for vision-language multi-task learning. arXiv preprint arXi...

    2023 arXiv

  3. [11]

    Aditya Chinchure, Pushkar Shukla, Gaurav Bhatt, Kiri Salij, Kartik Hosanagar, Leonid Sigal, and Matthew Turk. 2024. Tibet: Identifying and evaluating biases in text-to-image generative models. In European Conference on Computer Vision, pages 429--446. Springer

    2024

  4. [12]

    Jaemin Cho, Abhay Zala, and Mohit Bansal. 2023. Dall-eval: Probing the reasoning skills and social biases of text-to-image generation models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3043--3054

    2023

  5. [13]

    Kimberle Crenshaw. 1989. Demarginalizing the intersection of race and sex: A black feminist critique of antidiscrimination doctrine, feminist theory, and antiracist politics. In University of Chicago Legal Forum, pages 139--167

    1989

  6. [14]

    Tommy J Curry. 2018. Killing boogeymen: Phallicism and the misandric mischaracterizations of black males in theory. Res Philosophica

    2018

  7. [15]

    o rklund, and Henrik Bj \

    Hannah Devinney, Jenny Bj \"o rklund, and Henrik Bj \"o rklund. 2024. We don’t talk about that: case studies on intersectional analysis of social bias in large language models. In Workshop on Gender Bias in Natural Language Processing (GeBNLP), Bangkok, Thailand, 16th August, ...

    2024

  8. [16]

    Emily Diana, Wesley Gill, Michael Kearns, Krishnaram Kenthapadi, and Aaron Roth. 2021. Minimax group fairness: Algorithms and experiments. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, pages 66--76

    2021

  9. [17]

    Emily Diana and Alexander Williams Tolbert. 2023. Correcting underrepresentation and intersectional bias for classification. arXiv preprint arXiv:2306.11112

    2023 arXiv

  10. [18]

    Moreno D'Inc \`a , Elia Peruzzo, Massimiliano Mancini, Dejia Xu, Vidit Goel, Xingqian Xu, Zhangyang Wang, Humphrey Shi, and Nicu Sebe. 2024. Openbias: Open-set bias detection in text-to-image generative models. In Proceedings of the IEEE/CVF Conference on Computer Vision and P...

    2024

  11. [19]

    Piero Esposito, Parmida Atighehchian, Anastasis Germanidis, and Deepti Ghadiyaram. 2023. Mitigating stereotypical biases in text to image generative systems. arXiv preprint arXiv:2310.06904

    2023 arXiv

  12. [20]

    James R Foulds, Rashidul Islam, Kamrun Naher Keya, and Shimei Pan. 2020. An intersectional definition of fairness. In 2020 IEEE 36th international conference on data engineering (ICDE), pages 1918--1921. IEEE

    2020

  13. [21]

    Felix Friedrich, Patrick Schramowski, Manuel Brack, Lukas Struppek, Dominik Hintersdorf, Sasha Luccioni, and Kristian Kersting. 2023. Fair diffusion: Instructing text-to-image generation models on fairness. arXiv preprint arXiv:2302.10893

    2023 arXiv

  14. [22]

    Ismael Garrido-Mu \ n oz, Arturo Montejo-R \'a ez, Fernando Mart \' nez-Santiago, and L Alfonso Ure \ n a-L \'o pez. 2021. A survey on bias in deep nlp. Applied Sciences, 11(7):3184

    2021

  15. [23]

    Avijit Ghosh, Lea Genuit, and Mary Reagan. 2021. Characterizing intersectional group fairness with worst-case comparisons. In Artificial Intelligence Diversity, Belonging, Equity, and Inclusion, pages 22--34. PMLR

    2021

  16. [24]

    Sourojit Ghosh and Aylin Caliskan. 2023. ‘person’== light-skinned, western man, and sexualization of women of color: Stereotypes in stable diffusion. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 6971--6985

    2023

  17. [25]

    Wei Guo and Aylin Caliskan. 2021. Detecting emergent intersectional biases: Contextualized word embeddings contain a distribution of human-like biases. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, pages 122--133

    2021

  18. [26]

    Kimia Hamidieh, Haoran Zhang, Thomas Hartvigsen, and Marzyeh Ghassemi. 2023. Identifying implicit social biases in vision-language models

    2023

  19. [27]

    Lisa Anne Hendricks, Kaylee Burns, Kate Saenko, Trevor Darrell, and Anna Rohrbach. 2018. Women also snowboard: Overcoming bias in captioning models. In Proceedings of the European conference on computer vision (ECCV), pages 771--787

    2018

  20. [28]

    Yusuke Hirota, Yuta Nakashima, and Noa Garcia. 2022. Gender and racial bias in visual question answering datasets. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, pages 1280--1292

    2022

  21. [29]

    Elizabeth Hoepfinger. 2023. Racial and intersectional debiasing of contrastive language image pretraining. Master's thesis, University of Georgia

    2023

  22. [30]

    Phillip Howard, Avinash Madasu, Tiep Le, Gustavo Lujan Moreno, Anahita Bhiwandiwalla, and Vasudev Lal. 2024. Socialcounterfactuals: Probing and mitigating intersectional social biases in vision-language models with counterfactual examples. In Proceedings of the IEEE/CVF Confer...

    2024

  23. [31]

    Ben Hutchinson, Vinodkumar Prabhakaran, Emily Denton, Kellie Webster, Yu Zhong, and Stephen Denuyl. 2020. Social biases in nlp models as barriers for persons with disabilities. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5491--5501

    2020

  24. [32]

    Aparna R Joshi, Xavier Suau Cuadros, Nivedha Sivakumar, Luca Zappella, and Nicholas Apostoloff. 2022. Fair sa: Sensitivity analysis for fairness in face recognition. In Algorithmic fairness through the lens of causality and robustness workshop, pages 40--58. PMLR

    2022

  25. [33]

    Loukas Kavouras, Konstantinos Tsopelas, Giorgos Giannopoulos, Dimitris Sacharidis, Eleni Psaroudaki, Nikolaos Theologitis, Dimitrios Rontogiannis, Dimitris Fotakis, and Ioannis Emiris. 2023. Fairness aware counterfactuals for subgroups. Advances in Neural Information Processin...

    2023

  26. [34]

    Michael Kearns, Seth Neel, Aaron Roth, and Zhiwei Steven Wu. 2018. Preventing fairness gerrymandering: Auditing and learning for subgroup fairness. In International conference on machine learning, pages 2564--2572. PMLR

    2018

  27. [35]

    Dreyer, Aleksandar Shtedritski, and Yuki M

    Hannah Kirk, Yennie Jun, Haider Iqbal, Elias Benussi, Filippo Volpin, Frederic A. Dreyer, Aleksandar Shtedritski, and Yuki M. Asano. 2021. https://doi.org/10.48550/arXiv.2102.04130 Bias Out -of-the- Box : An Empirical Analysis of Intersectional Occupational Biases in Popular G...

    2021 doi arXiv

  28. [36]

    John P Lalor, Yi Yang, Kendall Smith, Nicole Forsgren, and Ahmed Abbasi. 2022. Benchmarking intersectional biases in nlp. In Proceedings of the 2022 conference of the North American chapter of the association for computational linguistics: Human language technologies, pages 3598--3609

    2022

  29. [37]

    Ida Marie S Lassen, Mina Almasi, Kenneth Enevoldsen, and Ross Deans Kristensen-McLachlan. 2023. Detecting intersectionality in ner models: A data-driven approach. In Proceedings of the 7th joint SIGHUM workshop on computational linguistics for cultural heritage, social science...

    2023

  30. [38]

    Daiqing Li, Aleks Kamko, Ehsan Akhgari, Ali Sabet, Linmiao Xu, and Suhail Doshi. 2024. https://arxiv.org/abs/2402.17245 Playground v2.5: Three insights towards enhancing aesthetic quality in text-to-image generation . Preprint, arXiv:2402.17245

    2024 arXiv

  31. [39]

    Bingyu Liu, Weihong Deng, Yaoyao Zhong, Mei Wang, Jiani Hu, Xunqiang Tao, and Yaohai Huang. 2019. Fair loss: Margin-aware reinforcement learning for deep face recognition. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10052--10061

    2019

  32. [40]

    Weicheng Ma, Brian Chiang, Tong Wu, Lili Wang, and Soroush Vosoughi. 2023. Intersectional stereotypes in large language models: Dataset and analysis. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 8589--8597

    2023

  33. [41]

    Nicole Meister, Dora Zhao, Angelina Wang, Vikram V Ramaswamy, Ruth Fong, and Olga Russakovsky. 2023. Gender artifacts in visual datasets. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4837--4848

    2023

  34. [42]

    Sungho Park, Sunhee Hwang, Jongkwang Hong, and Hyeran Byun. 2020. Fair-vqa: Fairness-aware visual question answering through sensitive attribute prediction. IEEE Access, 8:215091--215099

    2020

  35. [43]

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever. 2021. Zero-shot text-to-image generation. arXiv preprint arXiv:2102.12092

    2021 arXiv

  36. [44]

    Anton Razzhigaev, Arseniy Shakhmatov, Anastasia Maltseva, Vladimir Arkhipkin, Igor Pavlov, Ilya Ryabov, Angelina Kuts, Alexander Panchenko, Andrey Kuznetsov, and Denis Dimitrov. 2023. Kandinsky: An improved text-to-image synthesis with image prior and latent diffusion. In Proc...

    2023

  37. [45]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. 2022. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684--10695

    2022

  38. [46]

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Raphael Gontijo Lopes, and 1 others. 2022. Photorealistic text-to-image diffusion models with deep language understanding. arXiv...

    2022 arXiv

  39. [47]

    Preethi Seshadri, Sameer Singh, and Yanai Elazar. 2023. The bias amplification paradox in text-to-image generation. arXiv preprint arXiv:2308.00755

    2023 arXiv

  40. [48]

    Laleh Seyyed-Kalantari, Haoran Zhang, Matthew BA McDermott, Irene Y Chen, and Marzyeh Ghassemi. 2021. Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations. Nature medicine, 27(12):2176--2182

    2021

  41. [49]

    Deven Santosh Shah, H Andrew Schwartz, and Dirk Hovy. 2020. Predictive biases in natural language processing models: A conceptual framework and overview. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5248--5264

    2020

  42. [50]

    Arseniy Shakhmatov, Anton Razzhigaev, Aleksandr Nikolich, Vladimir Arkhipkin, Igor Pavlov, Andrey Kuznetsov, and Denis Dimitrov. 2023. kandinsky 2.2

    2023

  43. [51]

    Ryan Steed and Aylin Caliskan. 2021. Image representations learned with unsupervised pretraining contain human-like biases. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pages 701--713

    2021

  44. [52]

    Yi Chern Tan and L Elisa Celis. 2019. Assessing social and intersectional biases in contextualized word representations. Advances in neural information processing systems, 32

    2019

  45. [53]

    Angelina Wang, Alexander Liu, Ryan Zhang, Anat Kleiman, Leslie Kim, Dora Zhao, Iroha Shirai, Arvind Narayanan, and Olga Russakovsky. 2022 a . Revise: A tool for measuring and mitigating bias in visual datasets. International Journal of Computer Vision, 130(7):1790--1810

    2022

  46. [54]

    Angelina Wang, Vikram V Ramaswamy, and Olga Russakovsky. 2022 b . https://doi.org/10.1145/3531146.3533101 Towards Intersectionality in Machine Learning : Including More Identities , Handling Underrepresentation , and Performing Evaluation . In Proceedings of the 2022 ACM Confe...

    2022

  47. [55]

    Angelina Wang and Olga Russakovsky. 2023. Overwriting pretrained bias with finetuning data. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3957--3968

    2023

  48. [56]

    Jialu Wang, Xinyue Gabby Liu, Zonglin Di, Yang Liu, and Xin Wang. 2023 a . T2iat: Measuring valence and stereotypical biases in text-to-image generation. In Findings of the Association for Computational Linguistics: ACL 2023. Association for Computational Linguistics

    2023

  49. [57]

    Jialu Wang, Yang Liu, and Xin Wang. 2022 c . Assessing multilingual fairness in pre-trained multimodal representations. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2681--2695

    2022

  50. [58]

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. 2023 b . Exploring clip for assessing the look and feel of images. In AAAI

    2023

  51. [59]

    Zeyu Wang, Klint Qinami, Ioannis Christos Karakozis, Kyle Genova, Prem Nair, Kenji Hata, and Olga Russakovsky. 2020. Towards fairness in visual recognition: Effective strategies for bias mitigation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recog...

    2020

  52. [60]

    Cheng Zhang, Xuanbai Chen, Siqi Chai, Chen Henry Wu, Dmitry Lagun, Thabo Beeler, and Fernando De la Torre. 2023. Iti-gen: Inclusive text-to-image generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3969--3980

    2023

  53. [61]

    Eric Zhao, De-An Huang, Hao Liu, Zhiding Yu, Anqi Liu, Olga Russakovsky, and Anima Anandkumar. 2021. Scaling fair learning to hundreds of intersectional groups

    2021

  54. [62]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  55. [63]

    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 gl...

Pith tools

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