Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

More is Less? A Simulation-Based Approach to Dynamic Interactions between Biases in Multimodal Models

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper claims that whether a multimodal model amplifies, mitigates, or stays neutral on bias can be read off by comparing the fused bias score against each modality's score.

desk verdict A plausible taxonomy idea buried under an unreproducible simulation: the reported percentages are determined by unspecified formulas and contradicted by the paper's own counts. read the letter →

arxiv 2412.17505 v1 pith:4PTTAR2N submitted 2024-12-23 stat.ML cs.LG

classification stat.MLcs.LG
keywords multimodalbiasamplificationmitigationinteractionclassificationMMBiasdatasetcosinesimilaritysimulation-basedheuristicfairnessinAI
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 tries to establish that bias in multimodal models is not simply inherited from text and image inputs but arises from their interaction, and that this interaction can be classified into three types—amplification, mitigation, and neutrality—by comparing the multimodal bias score to the text-only and image-only scores. On the MMBias dataset, the author reports that the majority of interactions (67%) are neutral, 22% amplify bias, and 11% mitigate it, with mitigation occurring only when text bias is larger than image bias. The author argues that this simulation-based, threshold-comparison framework offers an interpretable way to audit multimodal systems and to detect when adding a modality helps, hurts, or does nothing to bias.

What carries the argument

The central object is the threshold-based comparison of three bias scores: text-only $S_t$, image-only $S_i$, and multimodal $S_m$, each computed as cosine similarity between embeddings and sentiment categories such as 'pleasant' versus 'unpleasant'. The classification rule $S_m > \max(S_t, S_i)$ (amplification), $S_m < \min(S_t, S_i)$ (mitigation), and $\min(S_t, S_i) \leq S_m \leq \max(S_t, S_i)$ (neutrality) carries the entire argument; $S_m$ is modeled as a noisy weighted combination of $S_t$ and $S_i$, so the counts of each interaction type are determined by that assumed fusion rule.

What would settle it

Compute $S_t$ and $S_i$ for the MMBias subcategories using the published dataset and any explicit embedding model, then fuse them with the paper's claimed weighted-plus-noise rule; if the resulting share of neutral interactions is not 67%, or if mitigation does not occur only under text dominance, the central empirical finding fails to reproduce. The paper does not provide enough detail to run this check, so the result is currently unfalsifiable from the text alone.

Watch

Extended reading notes

Core claim

The central claim is that the direction of bias interaction in a multimodal model is fully determined by the relative sizes of the unimodal bias scores $S_t$ (text) and $S_i$ (image) and the fused score $S_m$: amplification when $S_m > \max(S_t, S_i)$, mitigation when $S_m < \min(S_t, S_i)$, and neutrality in between. Applied to the MMBias dataset under the author's simulation-based bias scores, this rule yields 22% amplification, 11% mitigation, and 67% neutral interactions, with mitigation exclusively associated with text bias exceeding image bias and amplification equally likely under either modality's dominance.

Load-bearing premise

The analysis assumes that the undisclosed formulas used to compute $S_t$, $S_i$, and $S_m$, with $S_m$ as a noisy weighted combination of $S_t$ and $S_i$, are a valid representation of real multimodal bias; the reported percentages are fully determined by those formulas, so if a different fusion rule or embedding were used, the interaction counts would change.

Editorial extensions

If this is right

  • If the classification rule is accepted, bias interaction type can be audited from unimodal scores alone, making the framework applicable without re-training multimodal models.
  • The reported majority of neutral interactions implies that adding a modality usually does not automatically reduce bias, undercutting the assumption that triangulation across modalities is inherently debiasing.
  • The finding that mitigation occurs only under text dominance suggests text can counteract image bias when the text bias is the larger one, and that images play a stabilizing role in mitigation.
  • The paper recommends auditing single-modality bias before multimodal integration, and treating individual modality bias as a key intervention point.

Reading between the lines

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

  • Because the paper does not disclose the exact formulas for $S_t$, $S_i$, and $S_m$, or the fusion weight and noise scale, the reported 67/22/11 distribution is a property of the unstated simulation rather than of the MMBias data itself; changing the fusion rule or noise level would change the counts.
  • Under a convex-combination fusion rule with zero noise, all interactions would be neutral; the reported shares of amplification and mitigation are therefore artifacts of the noise term. A natural extension would be to vary the fusion weight and noise amplitude and plot how the interaction-type shares shift.
  • The framework's comparative logic is generic: the same $S_t$/$S_i$/$S_m$ comparison could be applied to audio-text or video-text embeddings, giving a direct test of whether text-dominance in mitigation replicates across modalities.
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 / 4 minor

Summary. The paper proposes a simulation-based heuristic framework for classifying interactions between text, image, and multimodal bias in the MMBias dataset. Bias scores for text (St), image (Si), and multimodal (Sm) embeddings are described qualitatively as derived from cosine similarities and a noisy weighted fusion. Interactions are classified by comparing Sm to St and Si into amplification (Sm exceeds both), mitigation (Sm is lower than both), and neutrality (Sm lies between them). The paper reports that on MMBias, 67% of interactions are neutral, 22% amplification, and 11% mitigation, with conditional probabilities suggesting that mitigation occurs only under text dominance and that image scores dominate on average. The central empirical claims are counts and percentages of these order-based interaction types.

Significance. If the approach were fully specified and validated, a transparent, interpretable framework for characterizing multimodal bias interactions could be a useful addition to the fairness literature, particularly because the paper addresses an under-explored question about how unimodal biases combine. However, in its current form the paper provides no equations, model names, embedding functions, simulation distributions, parameters, random seeds, or code. The reported interaction percentages are therefore not reproducible, and the internal inconsistencies in the counts and subcategory lists mean the empirical output cannot be checked against the paper's own data description. The paper's main strength is its framing of interaction types as a taxonomy and its attention to modality dominance, but this conceptual contribution is not enough to support the quantitative findings as stated.

major comments (4)
  1. [§3.3, §3.4, and Introduction] The bias scores St, Si, and Sm are never defined by equations, model names, embedding functions, or simulation parameters. The Introduction states that Sm is a weighted combination of text and image biases with additional noise, but no formula is given, and §3.3 provides only qualitative descriptions of cosine similarity computations. Because the classification in §3.2 is purely an ordering comparison among Sm, St, and Si, every reported percentage in §4.2 and §4.3 is determined by these undisclosed choices. The central empirical claims (22% amplification, 11% mitigation, 67% neutral, and the conditional probabilities) cannot be reproduced or independently verified without this information.
  2. [§3.1 and §4.1–§4.2] There are internal inconsistencies between the dataset description and the reported results. Section 3.1 lists 14 target groups, yet §4.2 reports 4 amplification + 1 mitigation + 13 neutral = 18 interactions. Additionally, §4.1 reports a minimum multimodal bias for the subcategory "French," but §3.1 lists only American, Arab, Chinese, and Mexican as nationality groups. The reported percentages also do not match the counts: 4/18, 1/18, and 13/18 equal 22.2%, 5.6%, and 72.2%, not the stated 22%, 11%, and 67%; using 14 groups gives 28.6%, 7.1%, and 64.3%. These discrepancies mean the quantitative findings are not anchored to the paper's own data description.
  3. [§3.5 and §4.3] The conditional probabilities in §3.5 are defined using counts under modality dominance, but the paper does not report the underlying counts of text-dominant versus image-dominant cases for the 14 or 18 subcategories. Consequently, the claims that amplification is split evenly (50/50) between text and image dominance, that mitigation occurs in 100% of cases under text dominance, and that neutral interactions are about 77% image-dominant cannot be checked. The text in §4.3 also notes that mitigation exhibits the highest average bias scores, which is counterintuitive to the definition of mitigation, and the suggested explanation of threshold overlap is not supported by any additional analysis.
  4. [§3.2 and §4.2] The interaction taxonomy is defined entirely by arithmetic comparisons of Sm with St and Si, and Sm is stated to be a noisy weighted combination of St and Si. As a result, the distribution of interaction types is largely a property of the assumed generative model rather than an empirical measurement of actual multimodal model behavior. The paper does not compare the simulated Sm values against outputs of any concrete text, image, or multimodal model on MMBias, nor does it validate the simulation against observed bias scores. Without such a comparison or a full specification of the generative model, the reported percentages should be interpreted as consequences of simulation assumptions, not as empirical findings about multimodal bias interactions in the dataset.
minor comments (4)
  1. [Abstract and §3.4] The abstract contains the typo "proportional analyzes," and §3.4 has "three 3 main categories." These should be corrected.
  2. [Figures 2–4] The figures are labeled only as "Author's compilation." They should include axis labels, units, and a description of how the plotted quantities are computed; currently the captions do not allow the reader to interpret the figures independently.
  3. [References] The reference list appears to contain two entries for Pena et al. 2023 with different author lists ("A. Pena, I. Serna, A. Morales, and J. Fierrez" and "Alejandro Peña, Ignacio Serna, Aythami Morales, Julian Fierrez, Alfonso Ortega, Ainhoa Herrarte, Manuel Alcantara, and Javier Ortega-Garcia"). These should be merged or disambiguated.
  4. [§3.3] The phrase "sentiment categories (e.g., 'pleasant' vs. 'unpleasant')" is vague; the paper should specify the exact word lists or embedding sources used for the cosine similarity computations.

Circularity Check

2 steps flagged · score 8.0 of 10

Central interaction-type distribution is forced by the paper's own definitions: Sm is a weighted combination of St and Si, and the taxonomy is purely an order comparison of Sm to St and Si.

  1. self definitional [Section 3.2 (Bias Dynamics' Definition) and Section 1 (Introduction)]
    "Bias amplification: Amplification occurs when the multimodal bias score Sm exceeds the larger of the two unimodal biases St for text-only bias, and Si for image-only bias Sm > max(St, Si). ... Multimodal bias is computed as a weighted combination of text and image biases, with additional noise to mimic real-world variability."

    The interaction taxonomy is defined entirely by the position of Sm relative to St and Si, while Sm is itself defined as a weighted combination of St and Si plus noise. Therefore every 'amplification', 'mitigation', or 'neutral' case is a logical consequence of the generative formula rather than an empirical property of the MMBias data. The reported counts simply enumerate regions of the (St, Si, noise) space under the chosen weights; they cannot confirm or disconfirm any hypothesis about real multimodal model behavior.

  2. fitted input called prediction [Abstract and Section 4.2/4.3]
    "The findings highlight that amplification (22%) occurs when text and image biases are comparable, while mitigation (11%) arises under the dominance of text bias, highlighting the stabilizing role of image bias. Neutral interactions (67%) are related to a higher text bias without divergence."

    The percentages 22%, 11%, and 67% are the direct output of applying the Section 3.2 threshold comparison to scores that are simulated as St, Si, and Sm = weighted combination of St and Si plus noise. The stated 'findings' about when amplification or mitigation occur are restatements of the defining inequalities Sm > max(St, Si) and Sm < min(St, Si) under the chosen generative model; they are not independent discoveries. Because the simulation parameters and embedding formulas are undisclosed, the numbers cannot be traced, but their dependence on the assumed construction is explicit and the reported statistics are forced by that construction.

full rationale

The paper's central empirical claim is the distribution of multimodal bias interaction types: 67% neutral, 22% amplification, 11% mitigation, together with conditional probabilities by modality dominance. The classification in Section 3.2 is a pure order comparison: amplification iff Sm > max(St, Si), mitigation iff Sm < min(St, Si), neutrality otherwise. Meanwhile, the Introduction states that bias scores were 'simulated probabilistically using controlled random sampling' and that 'multimodal bias is computed as a weighted combination of text and image biases, with additional noise.' Section 3.3 gives only qualitative descriptions ('cosine similarity between textual embeddings and sentiment categories') without any model names, embedding functions, fusion rule, noise distribution, or simulation parameters. Thus the percentages and conditional probabilities are determined by the definitions and the undisclosed generative model, not by an independently measurable property of the MMBias dataset. The paper does not compare against any external benchmark, real model, or held-out data; the conclusion itself calls the approach 'probabilistic simulation and simplistic modelling.' There is no self-citation load-bearing step, but the central claim reduces, by construction, to the assumed formula for Sm and the threshold definitions. This warrants a circularity score of 8: the result is forced by the paper's own definitions and generative assumptions, rather than being an empirical finding about multimodal bias dynamics.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The ledger is dominated by undisclosed simulation parameters and strong domain assumptions. Because the computation of St, Si, Sm is unspecified, the simulation parameters are effectively free parameters that determine the central results. No new entities are introduced.

free parameters (2)
  • Simulation parameters for text bias St and image bias Si (distributions, ranges, correlation) = unspecified
    The paper says bias scores were 'simulated probabilistically using controlled random sampling' (Introduction) but never specifies the distributions, so the generated scores and hence all interaction percentages depend on undisclosed choices.
  • Multimodal combination weight w and noise variance = unspecified
    Multimodal bias is described as 'a weighted combination of text and image biases, with additional noise' (Introduction), but the weight and noise level are not given. These choices directly determine which subcategories land in amplification, mitigation, or neutrality.
assumptions (4)
  • domain assumption Cosine similarity between embeddings and sentiment words is a valid measure of bias.
    Invoked in Section 3.3 for text and image bias scores without justification or validation.
  • ad hoc to paper Multimodal bias can be represented as a weighted sum of unimodal biases plus noise.
    Stated in the Introduction as the simulation model; this is an assumption specific to the paper's synthetic approach.
  • domain assumption The MMBias dataset and its 14 target groups are sufficient for drawing general conclusions about multimodal bias dynamics.
    Used in Section 3.1; no power analysis or evidence of representativeness is provided.
  • ad hoc to paper The threshold comparison (Sm > max(St,Si), etc.) captures meaningful interaction dynamics.
    Section 3.2 defines the framework; the meaningfulness of these categories is asserted, not derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of More is Less? A Simulation-Based Approach to Dynamic Interactions between Biases in Multimodal Models." pith.science (2026). https://pith.science/paper/4PTTAR2N

@misc{pith2026241217505,
  author       = {Pith},
  title        = {Pith review of: More is Less? A Simulation-Based Approach to Dynamic Interactions between Biases in Multimodal Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4PTTAR2N}},
  note         = {Machine review of arXiv:2412.17505}
}
read the original abstract

Multimodal machine learning models, such as those that combine text and image modalities, are increasingly used in critical domains including public safety, security, and healthcare. However, these systems inherit biases from their single modalities. This study proposes a systemic framework for analyzing dynamic multimodal bias interactions. Using the MMBias dataset, which encompasses categories prone to bias such as religion, nationality, and sexual orientation, this study adopts a simulation-based heuristic approach to compute bias scores for text-only, image-only, and multimodal embeddings. A framework is developed to classify bias interactions as amplification (multimodal bias exceeds both unimodal biases), mitigation (multimodal bias is lower than both), and neutrality (multimodal bias lies between unimodal biases), with proportional analyzes conducted to identify the dominant mode and dynamics in these interactions. The findings highlight that amplification (22\%) occurs when text and image biases are comparable, while mitigation (11\%) arises under the dominance of text bias, highlighting the stabilizing role of image bias. Neutral interactions (67\%) are related to a higher text bias without divergence. Conditional probabilities highlight the text's dominance in mitigation and mixed contributions in neutral and amplification cases, underscoring complex modality interplay. In doing so, the study encourages the use of this heuristic, systemic, and interpretable framework to analyze multimodal bias interactions, providing insight into how intermodal biases dynamically interact, with practical applications for multimodal modeling and transferability to context-based datasets, all essential for developing fair and equitable AI models.

Figures

Figures reproduced from arXiv: 2412.17505 by the authors.

Figure 1
Figure 1. Sample images from the MMBias dataset. Each row corresponds to one of the target classes: religion, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Bias Scores Across Categories (Author’s compilation) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Proportion of Cases by Text vs Image Dominance [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Order Matters: LVLMs as Judges for Temporal Reasoning in Image Sequences

    cs.CV 2026-08 conditional novelty 6.0 of 10

    LVLM judges are near chance when asked to pick the correctly ordered version of an image sequence, and this temporal blindness persists after fine-tuning and at larger scale.

Reference graph

Works this paper leans on

50 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [1]

    Vision-Language Models for Medical Report Generation and Visual Question Answering: A Review

    Iryna Hartsock and Ghulam Rasool. Vision-Language Models for Medical Report Generation and Visual Question Answering: A Review . arXiv.org, 2024. doi:10.48550/arXiv.2403.02469

  2. [2]

    Multimodal Categorization of Crisis Events in Social Media

    Mahdi Abavisani, Liwei Wu, Shengli Hu, Joel Tetreault, and Alejandro Jaimes. Multimodal Categorization of Crisis Events in Social Media . In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14667--14677, Dec 2020. doi:10.48550/arXiv.2004.04917

  3. [3]

    Analysis of Social Media Data using Multimodal Deep Learning for Disaster Response

    Ferda Ofli, Firoj Alam, and Muhammad Imran. Analysis of Social Media Data using Multimodal Deep Learning for Disaster Response . In International Conference on Information Systems for Crisis Response and Management, 2020. doi:10.48550/arXiv.2004.11838

  4. [4]

    Retrieval From and Understanding of Large-Scale Multi-modal Medical Datasets: A Review

    Henning M \"u ller and Devrim Unay. Retrieval From and Understanding of Large-Scale Multi-modal Medical Datasets: A Review . IEEE Transactions on Multimedia, 19 0 (9): 0 2093--2104, Dec. 2017. doi:10.1109/TMM.2017.2729400

  5. [5]

    CLIPSyntel: CLIP and LLM Synergy for Multimodal Question Summarization in Healthcare

    Akash Ghosh, Arkadeep Acharya, Raghav Jain, Sriparna Saha, Aman Chadha, and Setu Sinha. CLIPSyntel: CLIP and LLM Synergy for Multimodal Question Summarization in Healthcare . In AAAI Conference on Artificial Intelligence, volume 38, pages 22031--22039, 2023. doi:10.48550/arXiv.2312.11541

  6. [6]

    Beyond Medical Imaging - A Review of Multimodal Deep Learning in Radiology

    Lars Heiliger, Anjany Sekuboyina, Bjoern Menze, Jan Egger, and Jens Kleesiek. Beyond Medical Imaging - A Review of Multimodal Deep Learning in Radiology . TechRxiv, 19103432, 2022. doi:10.36227/techrxiv.19103432.v1

  7. [7]

    Using AI and Social Media Multimodal Content for Disaster Response and Management: Opportunities, Challenges, and Future Directions

    Muhammad Imran, Ferda Ofli, Doina Caragea, and Antonio Torralba. Using AI and Social Media Multimodal Content for Disaster Response and Management: Opportunities, Challenges, and Future Directions . Information Processing & Management, 57 0 (5): 0 102261, Dec. 2020. doi:10.1016/j.ipm.2020.102261

  8. [8]

    Fusing Visual and Textual Information to Determine Content Safety

    Rodrigo Leonardo, Amber Hu, Mohammad Uzair, Qiujing Lu, Iris Fu, Keishin Nishiyama, Sooraj Mangalath Subrahmannian, and Divyaa Ravichandran. Fusing Visual and Textual Information to Determine Content Safety . In 2019 18th IEEE International Conference On Machine Learning And Applications (ICMLA), pages 2026--2031, Dec 2019. doi:10.1109/ICMLA.2019.00324

Show all 50 references
  1. [9]

    Safety and Fairness for Content Moderation in Generative Models

    Susan Hao, Piyush Kumar, Sarah Laszlo, Shivani Poddar, Bhaktipriya Radharapu, and Renee Shelby. Safety and Fairness for Content Moderation in Generative Models . arXiv.org, 2023. doi:10.48550/arXiv.2306.06135

  2. [10]

    Formalizing Multimedia Recommendation through Multimodal Deep Learning

    Daniele Malitesta, Giandomenico Cornacchia, Claudio Pomo, Felice Antonio Merra, Tommaso Di Noia, and Eugenio Di Sciascio. Formalizing Multimedia Recommendation through Multimodal Deep Learning . ACM Transactions on Recommender Systems, 2023. doi:10.48550/arXiv.2309.05273

  3. [11]

    Review of multimodal machine learning approaches in healthcare

    Felix Krones, Umar Marikkar, Guy Parsons, Adam Szmul, and Adam Mahdi. Review of multimodal machine learning approaches in healthcare. arXiv.org, 2024. doi:10.48550/arXiv.2402.02460

  4. [12]

    Multi-modality approaches for medical support systems: A systematic review of the last decade

    Massimo Salvi, Hui Wen Loh, Silvia Seoni, Prabal Datta Barua, Salvador Garc \' a, Filippo Molinari, and U Rajendra Acharya. Multi-modality approaches for medical support systems: A systematic review of the last decade. Information Fusion, 103: 0 102134, Dec. 2024. doi:10.1016/...

  5. [13]

    Multimodal machine learning in precision health: A scoping review

    Adrienne Kline, Hanyin Wang, Yikuan Li, Saya Dennis, Meghan Hutch, Zhenxing Xu, Fei Wang, Feixiong Cheng, and Yuan Luo. Multimodal machine learning in precision health: A scoping review. NPJ Digital Medicine, 2022. doi:10.1038/s41746-022-00712-8

  6. [14]

    Performance evaluation of multimodal deep learning: object identification using UAV dataset

    Mingju He, Myron Hohil, Thomas LaPeruta, Kerolos Nashed, Victor Lawrence, and Yu-Dong Yao. Performance evaluation of multimodal deep learning: object identification using UAV dataset . In SPIE Proceedings, volume 11746, pages 602--608, Apr. 2021. doi:10.1117/12.2587825

  7. [15]

    Rethinking Multimodal Content Moderation from an Asymmetric Angle with Mixed-modality

    Jialin Yuan, Ye Yu, Gaurav Mittal, Matthew Hall, Sandra Sajeev, and Mei Chen. Rethinking Multimodal Content Moderation from an Asymmetric Angle with Mixed-modality . In Proceedings - 2024 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 8517--8527, May 2...

  8. [16]

    Robust Person Re-identification with Multi-Modal Joint Defence

    Yunpeng Gong, Liqing Huang, and Lifei Chen. Robust Person Re-identification with Multi-Modal Joint Defence . In IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, volume 2022-June, pages 4312--4321, Nov. 2021. doi:10.1109/CVPRW56347.2022.00477

  9. [17]

    Understanding public opinions on chinese short video platform by multimodal sentiment analysis using deep learning-based techniques

    Wei Shi, Jing Zhang, and Shaoyi He. Understanding public opinions on chinese short video platform by multimodal sentiment analysis using deep learning-based techniques. Kybernetes, ahead-of-print 0 (ahead-of-print), 2023. doi:10.1108/K-04-2023-0723/FULL/XML

  10. [18]

    MAMSC: a semantic enhanced representation model for public opinion key node recognition based on multianchor mapping in semantic communities

    Yongcong Luo and He Zhu. MAMSC: a semantic enhanced representation model for public opinion key node recognition based on multianchor mapping in semantic communities . Kybernetes, ahead-of-print 0 (ahead-of-print), 2024. doi:10.1108/K-04-2024-0914/FULL/XML

  11. [19]

    Detection and Evaluation of Machine Learning Bias

    Salem Alelyani. Detection and Evaluation of Machine Learning Bias . Applied Sciences, 11 0 (14): 0 6271, Jul. 2021. doi:10.3390/APP11146271

  12. [20]

    Bias in Machine Learning: A Literature Review

    Konstantinos Mavrogiorgos, Athanasios Kiourtis, Argyro Mavrogiorgou, Andreas Menychtas, and Dimosthenis Kyriazis. Bias in Machine Learning: A Literature Review . Applied Sciences, 14 0 (19): 0 8860, Oct. 2024. doi:10.3390/APP14198860

  13. [21]

    A Survey on Bias and Fairness in Machine Learning

    Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. A Survey on Bias and Fairness in Machine Learning . ACM Computing Surveys (CSUR), 54 0 (6), July 2021. doi:10.1145/3457607

  14. [22]

    RUBi: Reducing Unimodal Biases in Visual Question Answering

    Remi Cadene, Corentin Dancette, Matthieu Cord, Devi Parikh, et al. RUBi: Reducing Unimodal Biases in Visual Question Answering . In Neural Information Processing Systems, 2019. doi:10.48550/arXiv.1906.10169

  15. [23]

    Towards Understanding and Mitigating Social Biases in Language Models

    Paul Pu Liang, Chiyu Wu, Louis-Philippe Morency, and Ruslan Salakhutdinov. Towards Understanding and Mitigating Social Biases in Language Models . In International Conference on Machine Learning, 2021. doi:10.48550/arXiv.2106.13219

  16. [24]

    Biases in Large Language Models: Origins, Inventory, and Discussion

    Navigli, Roberto and Conia, Simone and Ross, Bj \"o rn . Biases in Large Language Models: Origins, Inventory, and Discussion . Journal of Data and Information Quality, 15 0 (2): 0 1--21, June 2023. doi:10.1145/3597307

  17. [25]

    Quantifying and Mitigating Unimodal Biases in Multimodal Large Language Models: A Causal Perspective

    Meiqi Chen, Yixin Cao, Yan Zhang, and Chaochao Lu. Quantifying and Mitigating Unimodal Biases in Multimodal Large Language Models: A Causal Perspective . arXiv.org, 2024. doi:10.48550/arXiv.2403.18346

  18. [26]

    K. Miller. Covert Racism in AI: How Language Models Are Reinforcing Outdated Stereotypes , 2024. URL https://hai.stanford.edu/news/covert-racism-ai-how-language-models-are-reinforcing-outdated-stereotypes. Accessed: Nov. 25, 2024

  19. [27]

    Kotek, R

    H. Kotek, R. Dockum, and D. Q. Sun. Gender bias and stereotypes in Large Language Models . In Proceedings of the ACM Collective Intelligence Conference, CI 2023, pages 12--24, November 2023. doi:10.1145/3582269.3615599

  20. [28]

    Manzini, L

    T. Manzini, L. Yao Chong, A. W. Black, and Y. Tsvetkov. Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings . In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguist...

  21. [29]

    Discover and Mitigate Multiple Biased Subgroups in Image Classifiers

    Zeliang Zhang, Mingqian Feng, Zhiheng Li, and Chenliang Xu. Discover and Mitigate Multiple Biased Subgroups in Image Classifiers . Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10906--10915, 2024. doi:10.1109/CVPR52733.2024.01037

  22. [30]

    Boury-Brisset and J

    A.-C. Boury-Brisset and J. Berger. Benefits and Challenges of AI/ML in Support of Intelligence and Targeting in Hybrid Military Operations . Technical report, NATO S&I Organisation, 2020. URL https://www.sto.nato.int/publications/STO Accessed: Oct. 04, 2024

  23. [31]

    B. M. Booth, L. Hickman, S. K. Subburaj, L. Tay, S. E. Woo, and S. K. D’Mello. Bias and fairness in multimodal machine learning: A case study of automated video interviews . In Proceedings of the 2021 international conference on multimodal interaction, pages 268--277, 2021. do...

  24. [32]

    A. Pena, I. Serna, A. Morales, and J. Fierrez. Bias in Multimodal AI: Testbed for Fair Automatic Recruitment . In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 129--137, Nov. 2023. doi:10.48550/arXiv.2004.07173

  25. [33]

    Y. Shen, H. Di, and S. M. Mitigating Biases in Multimodal Personality Assessment . In International Conference on Multimodal Interaction, 2020. doi:10.1145/3382507.3418889

  26. [34]

    Shengjia, R

    Z. Shengjia, R. Hongyu, Y. Arianna, S. Jiaming, D. G. Noah, and E. Stefano. Bias and Generalization in Deep Generative Models: An Empirical Study . Neural Information Processing Systems, 2018. doi:10.48550/arXiv.1811.03259

  27. [35]

    A Prompt Array Keeps the Bias Away: Debiasing Vision-Language Models with Adversarial Learning

    Hugo Berg, Siobhan Mackenzie Hall, Yash Bhalgat, Wonsuk Yang, Hannah Rose Kirk, Aleksandar Shtedritski, and Max Bain. A Prompt Array Keeps the Bias Away: Debiasing Vision-Language Models with Adversarial Learning . In AACL, 2022. doi:10.48550/arXiv.2203.11933

  28. [36]

    Towards Multimodal Sentiment Analysis Debiasing via Bias Purification

    Dingkang Yang, Mingcheng Li, Dongling Xiao, Yang Liu, Kun Yang, Zhaoyu Chen, Yuzheng Wang, Peng Zhai, Ke Li, and Lihua Zhang. Towards Multimodal Sentiment Analysis Debiasing via Bias Purification . arXiv.org, 2024. doi:10.48550/arXiv.2403.05023

  29. [37]

    Alasadi, R

    J. Alasadi, R. Arunachalam, P. K. Atrey, and V. K. Singh. A Fairness-Aware Fusion Framework for Multimodal Cyberbullying Detection . In 2020 IEEE Sixth International Conference on Multimedia Big Data (BigMM), pages 166--173, Dec. 2020. doi:10.1109/BigMM50055.2020.00032

  30. [39]

    Acosta, N

    H. Acosta, N. Henderson, J. Rowe, W. Min, J. Minogue, and J. Lester. What’s Fair is Fair: Detecting and Mitigating Encoded Bias in Multimodal Models of Museum Visitor Attention . In Proceedings of the 2021 International Conference on Multimodal Interaction, pages 258--267, Oct...

  31. [40]

    Mandal, S

    A. Mandal, S. Leavy, and S. Little. Measuring Bias in Multimodal Models: Multimodal Composite Association Score . In Communications in Computer and Information Science, pages 17--30. Springer Nature Switzerland, 2023 b . doi:10.48550/arXiv.2304.13855

  32. [41]

    Frontiers of multimodal learning: A responsible AI approach

    Microsoft Research Blog . Frontiers of multimodal learning: A responsible AI approach . https://www.microsoft.com/en-us/research/blog/frontiers-of-multimodal-learning-a-responsible-ai-approach/, 2024. Accessed: Oct. 25, 2024

  33. [42]

    Bhargava and D

    S. Bhargava and D. Forsyth. Exposing and Correcting the Gender Bias in Image Captioning Datasets and Models . arXiv preprint arXiv:1912.00578, 2019. doi:10.48550/arXiv.1912.00578

  34. [43]

    J. Zhao, T. Wang, M. Yatskar, V. Ordonez, and K. W. Chang. Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints . In EMNLP 2017 - Conference on Empirical Methods in Natural Language Processing, Proceedings, pages 2979--2989. Association for...

  35. [44]

    Janghorbani and G

    S. Janghorbani and G. de Melo. MultiModal Bias: Introducing a Framework for Stereotypical Bias Assessment beyond Gender and Race in Vision Language Models . In Conference of the European Chapter of the Association for Computational Linguistics, 2023. doi:10.48550/arXiv.2303.12734

  36. [45]

    Sawhney, A

    R. Sawhney, A. Aggarwal, and R. R. Shah. An Empirical Investigation of Bias in the Multimodal Analysis of Financial Earnings Calls . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologie...

  37. [46]

    R. Wang, P. Chaudhari, and C. Davatzikos. Bias in machine learning models can be significantly mitigated by careful training: Evidence from neuroimaging studies . Proceedings of the National Academy of Sciences, 120 0 (6), 2023. doi:10.1073/pnas.2211613120

  38. [47]

    Schmitz, R

    M. Schmitz, R. Ahmed, and J. Cao. Bias and Fairness on Multimodal Emotion Detection Algorithms . arXiv.org, 2022. doi:10.48550/arXiv.2205.08383

  39. [48]

    Alabdulmohsin, X

    I. Alabdulmohsin, X. Wang, A. Steiner, P. Goyal, A. D'Amour, and X. Zhai. CLIP the Bias: How Useful is Balancing Data in Multimodal Learning? ArXiv, 2024. doi:10.48550/arXiv.2403.04547

  40. [49]

    Srinivasan and Y

    T. Srinivasan and Y. Bisk. Worst of Both Worlds: Biases Compound in Pre-trained Vision-and-Language Models . In Proceedings of the 4th Workshop on Gender Bias in Natural Language Processing (GeBNLP), 2022. doi:10.18653/v1/2022.gebnlp-1.10

  41. [50]

    Human-Centric Multimodal Machine Learning: Recent Advances and Testbed on AI-Based Recruitment

    Alejandro Pe \ n a, Ignacio Serna, Aythami Morales, Julian Fierrez, Alfonso Ortega, Ainhoa Herrarte, Manuel Alcantara, and Javier Ortega-Garcia. Human-Centric Multimodal Machine Learning: Recent Advances and Testbed on AI-Based Recruitment . SN Comput Sci, 2023. doi:10.1007/s4...

  42. [51]

    P. Kok, J. F. M. Jehee, and F. P. de Lange. Less Is More: Expectation Sharpens Representations in the Primary Visual Cortex . Neuron, 75 0 (2): 0 265--270, Jul 2012. doi:10.1016/J.NEURON.2012.04.034

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.