Assessing Intersectional Bias in Representations of Pre-Trained Image Recognition Models

Sebastian Stober; Valerie Krug

REVIEW 4 major objections 5 minor 40 references

Assessing Intersectional Bias in Representations of Pre-Trained Image Recognition Models

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 Pre-trained ImageNet classifiers, regardless of architecture, strongly encode age information in their representations of facial images, while race and gender signals are weaker.

desk verdict A transparent but overclaimed extension of the authors' own prior work: the age-differentiation result is plausible and consistent across architectures, but the headline 'strongly encode age' rests on in-sample probe accuracy the paper itself admits does not generalize. read the letter →

arxiv 2506.03664 v2 pith:3SUGTPVB submitted 2025-06-04 cs.CV cs.CYcs.HCcs.LG

Valerie Krug , Sebastian Stober This is my paper

classification cs.CVcs.CYcs.HCcs.LG

keywords intersectionalbiaspre-trainedimageclassifiersrepresentationlinearclassifierprobesNeuronActivationProfilestopographicmapsFairFaceage

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 asks whether commonly reused image-recognition backbones carry demographic bias inside their learned representations, not just in their final predictions. Using linear classifier probes and topographic visualizations of neuron activations on the FairFace face dataset, it finds that VGG16, ResNet50, and InceptionV3 all encode age strongly across their middle layers, so much that age groups are the easiest demographic division to recover linearly. Race is present but weaker, mostly in confusions among certain ethnicities, and gender is mostly separable only in middle-aged groups. The authors conclude that transfer-learning users inherit an age signal in the feature extractor regardless of architecture, and that linear probes alone underestimate how usable the bias is because non-linear classifiers would likely exploit it.

What carries the argument

The argument is carried by two instruments: linear classifier probes, which train a linear readout on a layer's activations to predict the 126 intersectional groups and treat higher accuracy as evidence that the concept is encoded; and Neuron Activation Profiles (NAPs), which average activations per group, subtract the overall expected activation, and lay the resulting patterns out as topographic activation maps via a UMAP projection with a particle-repulsion layout. The probes give the quantitative separability curve across layers, while the maps give a visual counterpart in which similar group patterns predict which groups the model will confuse. A third, smaller mechanism is the error table analysis, which names the most frequent misclassifications and thereby locates the bias in particular intersections.

What would settle it

A decisive check would be to rerun the same probes on FairFace images whose demographic labels are kept but whose appearance correlates are removed or controlled, for example by normalizing skin texture to remove wrinkles or cropping out hair and clothing; if the age probe's accuracy collapses to chance while accuracy on the appearance controls stays high, the central claim is reduced to appearance encoding, and if the age accuracy survives such normalization, the claim is confirmed.

Watch

Extended reading notes

Core claim

The paper's central claim is that pre-trained ImageNet classifiers, regardless of architecture, strongly encode age information in their representations, with less pronounced encoding of race and gender. Across three architectures, central layers' activations allow a linear probe to separate the 126 intersectional groups in training data, but validation accuracy stays low, meaning the signal is real yet not general enough for a linear readout. Error analysis and topographic activation maps show age errors dominate and usually move to adjacent age groups; race errors concentrate on confusion with 'Latino_Hispanic', 'White', 'Indian', and 'Middle Eastern'; gender errors are rare and gender differences appear mainly for middle ages. The paper interprets this as evidence that the bias lives in the feature extractor itself and would be available to a non-linear classifier during transfer learning.

Load-bearing premise

The argument depends on treating FairFace's demographic labels as the thing the representations are encoding: if the models are actually encoding wrinkles, hairstyles, facial hair, or other appearance cues that merely line up with the labels, then the claimed 'age bias' is an observation about visual appearance rather than about a demographic concept.

Editorial extensions

If this is right

A downstream model built on any of these backbones can, in principle, read off age from the feature extractor, so age bias transfers even to tasks unrelated to age.
Because the age signal appears in all three architectures, switching backbone is not a remedy; mitigation would have to act on the representation itself.
Race bias shows up as systematic confusions among particular ethnicities, so auditing single-race accuracy is not enough.
Gender differences are concentrated in middle-aged groups, so gender-bias measurements that average across ages can hide the effect.
The gap between high training and low validation accuracy on the probes implies that a non-linear classifier would likely exploit the encoded bias more fully than a linear probe can.

Reading between the lines

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

The authors leave open whether the effect is specific to ImageNet; a natural extension is to run the same probe and NAP pipeline on models trained on non-face and face-specific datasets, which would separate data-source effects from architecture effects.
A quantitative pairwise similarity index derived from the NAP maps could turn the visual confusions into a per-intersection audit metric, letting practitioners flag the most confusable demographic cells before deployment.
If the age signal is largely carried by skin-texture and face-shape cues, then the same measurement could be re-framed as a test of whether representations encode 'facial age appearance' rather than social age; the paper's method cannot distinguish these, and that distinction matters for deciding whether intervention is needed.
The failure of linear probes to generalize suggests that representation-bias audits should treat linear separability as a lower bound and include non-linear probes or few-shot readouts to estimate the bias a real downstream model would experience.

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.

Desk Editor's Note

read the letter

Quick take: this is an incremental paper from a group that has been building the NAP/topographic-map toolset for a while. The new bit is applying it to intersections of race, age, and gender across three ImageNet CNNs, and adding linear probes. The empirical pattern — age differences dominate, race confusion mainly around Latino/Hispanic, gender effects in middle ages — is consistent across VGG16, ResNet50, and InceptionV3, and matches their earlier clustering study. The code is public and the methods are described clearly enough to reproduce.

What I like: the authors are unusually candid about the probes not generalizing. Section 7 says it outright. The topographic maps are a nice descriptive device, and the qualitative reading of adjacent-age similarity lines up with the error tables.

The soft spot is exactly where the stress-test note lands. The abstract and conclusion claim that ImageNet classifiers “strongly encode age information,” but the linear probe evidence is in-sample: training accuracy near 1.0, validation accuracy far lower, and no shuffled-label control. With 126 output classes and a fully connected probe that can have millions of parameters, memorization is a real possibility. The validation accuracy is above chance and the trends are consistent, so the result is not empty, but it supports “some linearly decodable age signal in the training distribution” rather than “strongly encode” in a generalizable sense. The topographic maps are computed on the same training data and could reflect correlated visual cues (wrinkles, hairstyle) rather than demographic concepts. To be fair, the authors do flag that linear probes alone are insufficient, but then they use them as the basis for the headline claim anyway. That tension is the paper's main problem.

Is it new? Only modestly. The age-dominance finding was already in their IJCNN clustering paper (ref 23). This adds multi-architecture and linear-probe evidence, which is useful confirmation but not a conceptual jump.

Bottom line: it's an honest, reproducible descriptive study useful to people working on bias in transfer learning, but the central claim needs to be dialed back to match the evidence. A serious referee could push them toward held-out evaluation or a more cautious interpretation. I'd send it to review, but I would not cite it as evidence of “strong encoding” without seeing a generalization check.

Referee Report

4 major / 5 minor

Summary. The paper investigates intersectional bias in the internal representations of three pre-trained ImageNet CNNs (VGG16, ResNet50, InceptionV3) using the FairFace dataset. For each of 126 intersectional groups defined by race, age, and gender, the authors train linear classifier probes on layer activations and visualize average activations as topographic maps. The central claim is that these models, regardless of architecture, 'strongly encode age information' in their representations, with weaker encoding of ethnicity and of gender in middle-aged groups.

Significance. If the central claim were established, the paper would be a useful contribution to the transfer-learning fairness literature, because it would identify a generalizable demographic signal in widely used backbone representations. The paper also has concrete strengths: the code is publicly released, the evaluation uses an independent and relatively balanced face dataset, and the intersectional grouping across 126 categories is more granular than much prior work. However, the main evidence offered for the central claim is in-sample linear-probe accuracy, which the authors themselves report as failing to generalize, and the qualitative topographic-map analysis is not validated against a quantitative baseline. As a result, the significance of the paper is currently limited by the gap between its conclusions and the strength of the evidence.

major comments (4)

[§6.4, §7, Figs. 5, 7, 9] The central claim that pre-trained ImageNet classifiers 'strongly encode age information' rests on training accuracy of linear probes, but the paper's own results show that validation accuracy is far lower and the authors state in Section 7 that the probes 'could distinguish training data, but did not generalize to unseen data.' In-sample separability by a high-capacity probe is not evidence of generalizable encoding; it is consistent with memorization or overfitting. No shuffled-label control or evaluation on FairFace's official held-out split is reported, so the abstract's conclusion is not supported by the probe experiments as presented.
[§5.2, §6.4] The argument that FairFace is an 'independent evaluation data set' because its labels are not ImageNet categories and its examples were not used in training does not rule out encoding of visual correlates of the labels. Age labels are strongly correlated with wrinkles, skin texture, hairstyle, facial shape, and other appearance cues; the paper itself says in Section 6.4 that age differences are 'driven by change of skin texture with age.' This undercuts the interpretation that the models encode 'age information' as a demographic concept rather than low-level visual attributes. The conclusion needs to either control for such correlates or be carefully reframed as a claim about visual appearance correlates.
[§5.3, Figs. 5, 7, 9] The probe setup is not calibrated against a null model. After downsampling activations to at most 8x8 spatial resolution, InceptionV3's mixed7 layer has 768 channels, yielding about 49,152 input features to a fully connected layer with 126 outputs, trained on roughly 14.5k examples (90% of 126 batches of up to 128). With this parameter count and class count, high training accuracy can be achieved even for random labels. The paper reports no shuffled-label baseline, no repeated runs with different random splits, and no confidence intervals. Therefore the near-1.0 training accuracies in Figures 5, 7, and 9 do not by themselves demonstrate that the representations 'strongly encode' the intersectional groups.
[§5.4, §6.4] The topographic-map interpretation relies on subjective visual comparison of average activation patterns, and the paper does not quantify the similarity or dissimilarity that drives the conclusions, such as 'activation differences between genders mostly emerge for middle ages' or 'age differences are most pronounced.' Without a quantitative similarity measure or a statistical test, these qualitative claims cannot be independently verified. They are used to reinforce the central claim and should either be backed by a quantitative analysis or be presented as hypotheses rather than findings.

minor comments (5)

[§1, §7] The phrase 'regardless of the architecture' is too strong given that only three CNN architectures, all trained on ImageNet with similar supervision, are examined; the paper itself lists Vision Transformers as future work.
[§5.3] The relationship between the 128-example batch per group used for probes and the 640-example cap in Figure 1b should be clarified; it is not clear whether the probe batches are drawn from the capped set or from the full set.
[§6.2] The text says 'High training accuracy with low performance on the validation accuracy indicates overfitting,' but the paper then uses training accuracy to rank layers and identify biases; this inconsistency should be addressed by stating explicitly which quantity is intended to measure encoding.
[§6.4] The statement that multi-layer non-linear classifiers 'would be more likely to learn generalized patterns from the activations' is speculative and not tested; it should be removed or supported by experiments.
[§6.3] There are typographical errors, including 'missclassfications' and 'classfication' in the classification-errors paragraph.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the central claim rests on independent linear probes and activation visualizations; the sole corroborating self-citation is not load-bearing.

full rationale

The paper contains no derivation chain in which an output quantity is defined from, fitted to, or forced by the quantity it claims to predict. The linear probes are an external method (Alain and Bengio [2]) applied to activations of externally pre-trained models on the externally provided FairFace data set; NAPs and topographic activation maps are re-specified in Section 4.2 rather than imported as unstated assumptions. The paper explicitly flags the main limitation on its own evidence: 'Linear classifier probes could distinguish training data, but did not generalize to unseen data' (Section 7). This is a generalizability threat and a possible gap between in-sample separability and the conclusion 'strongly encode age information,' but it is not a circular reduction: training and validation accuracies are reported as measured, no fitted parameter is renamed as a prediction, and no equation is shown to equal its own input. The only self-citation touching the central claim is the sentence 'This is in agreement with a recent related study that follows a clustering-based analysis [23],' where [23] is the authors' own IJCNN paper. That corroboration is not load-bearing because the conclusion is already drawn from the paper's own experiments, and earlier self-citations [20-22] are methodological lineage for NAPs and topographic maps, explicitly described in the methods section. Therefore no circular step is exhibited; the score of 2 reflects one minor non-load-bearing self-citation rather than any actual circularity.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new theoretical entities or fitted theoretical constants. Its free parameters are experimental design choices: the group-size cap, probe hyperparameters, and the post hoc selection of representative layers. The axioms are standard domain assumptions about probe validity, label accuracy, and visual interpretation that are not independently verified.

free parameters (3)

max_group_size_640 = 640
Groups capped at 640 examples to reduce imbalance (Section 5.2); chosen by hand, affects all subsequent counts and splits.
linear_probe_hyperparameters = dropout 0.25, L1 1e-5, L2 1e-4, 50 epochs
Fixed hyperparameters of the probe classifiers (Section 5.3); results depend on their choice, but they are not fitted to the data to produce the claim.
layer_selection = VGG16 layer 4, ResNet50 layer 15, InceptionV3 layer 9
Representative layers are chosen by highest probe accuracy (Sections 6.1 to 6.3), which is a post hoc selection that shapes the error analysis and visual conclusions.

assumptions (3)

domain assumption Linear classifier probes measure what information a layer encodes.
The paper assumes linear classifier probes are valid probes of representation content (Section 4.1), but the gap between training and validation accuracy shows the probe does not capture generalizable information.
domain assumption FairFace demographic labels are correct and the dataset is independent of ImageNet.
The analysis treats FairFace demographic labels as accurate ground truth and assumes independence from the models' training data (Section 5.2), but label noise or correlated visual cues are not controlled.
domain assumption Similar activation patterns across groups indicate likely confusion and hence bias.
The topographic map interpretation (Section 5.4) asserts that similar activation patterns indicate groups are hard to distinguish, a rule that is not validated against behavioral data.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Assessing Intersectional Bias in Representations of Pre-Trained Image Recognition Models." pith.science (2026). https://pith.science/paper/3SUGTPVB

@misc{pith2026250603664,
  author       = {Pith},
  title        = {Pith review of: Assessing Intersectional Bias in Representations of Pre-Trained Image Recognition Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3SUGTPVB}},
  note         = {Machine review of arXiv:2506.03664}
}

read the original abstract

Deep Learning models have achieved remarkable success. Training them is often accelerated by building on top of pre-trained models which poses the risk of perpetuating encoded biases. Here, we investigate biases in the representations of commonly used ImageNet classifiers for facial images while considering intersections of sensitive variables age, race and gender. To assess the biases, we use linear classifier probes and visualize activations as topographic maps. We find that representations in ImageNet classifiers particularly allow differentiation between ages. Less strongly pronounced, the models appear to associate certain ethnicities and distinguish genders in middle-aged groups.

Figures

Figures reproduced from arXiv: 2506.03664 by the authors.

**Figure 2.** Exemplary accuracies along layers. 0 20 40 epoch 0.0 0.5 1.0 accuracy training validation [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 4.** Exemplary sets of topographic activation maps from the NAP analysis. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

Figures from the paper (6 more)

**Figure 5.** Figure 5: Linear classifier probe accuracies along the VGG16 layers and selected learning curves using [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

**Figure 6.** Figure 6: Topographic activation maps for all subgroups in VGG16 layer 4 (“block4_pool”). [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

**Figure 7.** Figure 7: Classifier probe accuracies along the ResNet50 layers and selected learning curves with [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

**Figure 8.** Figure 8: Topographic activation maps for all subgroups in ResNet50 layer 15 (“conv5_block1_out”). [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

**Figure 9.** Figure 9: Classifier probe accuracies along the InceptionV3 layers and selected learning curves with [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

**Figure 10.** Figure 10: Topographic activation maps for all subgroups in InceptionV3 layer 9 (“mixed7”). [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 30 canonical work pages

[1]

Ahn and A

J. Ahn and A. Oh. Mitigating language-dependent ethnic bias in BERT. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 533– 549, Online and Punta Cana, Dominican Republic, Nov. 2021. Association for Computational Linguistics

work page 2021
[2]

Alain and Y

G. Alain and Y. Bengio. Understanding intermediate layers using linear classifier probes. In International Conference on Learning Representations (ICLR), Workshop Track Proceedings, 2017

work page 2017
[3]

Baker, H

N. Baker, H. Lu, G. Erlikhman, and P. J. Kellman. Deep convolutional networks do not classify based on global object shape.PLoS computational biology, 14(12):e1006613, 2018

work page 2018
[4]

Bolukbasi, K.-W

T. Bolukbasi, K.-W. Chang, J. Y. Zou, V. Saligrama, and A. T. Kalai. Man is to computer programmer as woman is to homemaker? Debiasing Word Embeddings. InAdvances in Neural Information Processing Systems, volume 29, pages 4349–4357, 2016

work page 2016
[5]

Buolamwini and T

J. Buolamwini and T. Gebru. Gender Shades: Intersectional Accuracy Disparities in Commer- cial Gender Classification. InProceedings of the 1st Conference on Fairness, Accountability and Transparency, volume 81, pages 77–91. PMLR, 23–24 Feb 2018

work page 2018
[6]

Carter, Z

S. Carter, Z. Armstrong, L. Schubert, I. Johnson, and C. Olah. Activation atlas.Distill, 4(3):e15, 2019

2019
[7]

J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In2009 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 248–255. Ieee, 2009

work page 2009
[8]

Devlin, M.-W

J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.arXiv preprint arXiv:1810.04805, 2019

arXiv 2019

Show all 40 references

[9]

Dosovitskiy, L

A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. De- hghani, M. Minderer, G. Heigold, S. Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

2010 arXiv
[10]

Erhan, Y

D. Erhan, Y. Bengio, A. Courville, and P. Vincent. Visualizing higher-layer features of a deep network.University of Montreal, 1341(3):1, 2009

2009
[11]

Fiacco, S

J. Fiacco, S. Choudhary, and C. Rose. Deep neural model inspection and comparison via func- tional neuron pathways. InAnnual Meeting of the Association for Computational Linguistics (ACL), pages 5754–5764, 2019

2019
[12]

Goodfellow, Y

I. Goodfellow, Y. Bengio, A. Courville, and Y. Bengio.Deep learning, volume 1. MIT press Cambridge, 2016

2016
[13]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770– 778, 2016

2016
[14]

Hohman, H

F. Hohman, H. Park, C. Robinson, and D. H. P. Chau. Summit: Scaling deep learning inter- pretability by visualizing activation and attribution summarizations.IEEE Transactions on Visualization and Computer Graphics, 26(1):1096–1106, 2019

2019
[15]

M. Hort, Z. Chen, J. M. Zhang, M. Harman, and F. Sarro. Bias mitigation for machine learning classifiers: A comprehensive survey.ACM Journal on Responsible Computing, 2023. Intersectional Bias in Pre-Trained Image Recognition Models 19

2023
[16]

Karkkainen and J

K. Karkkainen and J. Joo. FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1548–1558, 2021

2021
[17]

B. Kim, M. Wattenberg, J. Gilmer, C. Cai, J. Wexler, F. Viegas, et al. Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV). InProceedings of the International Conference on Machine Learning (ICML), pages 2668– 2677, 2018

2018
[18]

S. Kim, A. Gholami, A. Shaw, N. Lee, K. Mangalam, J. Malik, M. W. Mahoney, and K. Keutzer. Squeezeformer: An efficient transformer for automatic speech recognition. In Advances in Neural Information Processing Systems, volume 35, pages 9361–9373, 2022

2022
[19]

Kindermans, K

P.-J. Kindermans, K. T. Schütt, M. Alber, K.-R. Müller, D. Erhan, B. Kim, and S. Dähne. Learninghowtoexplainneuralnetworks:PatternNetandPatternAttribution. InInternational Conference on Learning Representations (ICLR), 2018

2018
[20]

V. Krug. Neuroscience-Inspired Analysis and Visualization of Deep Neural Networks, 2024

2024
[21]

V. Krug, C. Olson, and S. Stober. Visualizing Bias in Activations of Deep Neural Networks as Topographic Maps. InProceedings of the 1st Workshop on Fairness and Bias in AI (AE- QUITAS 2023) co-located with 26th European Conference on Artificial Intelligence (ECAI

2023
[22]

V. Krug, R. K. Ratul, C. Olson, and S. Stober. Visualizing Deep Neural Networks with Topographic Activation Maps. InHHAI 2023: Augmenting Human Intellect, pages 138–152. IOS Press, 2023

2023
[23]

V. Krug, F. Röhrbein, and S. Stober. Intersectional Bias Quantification in Facial Image Processing with Pre-Trained ImageNet Classifiers. In2025 International Joint Conference on Neural Networks (IJCNN), pages 1–8, 2025

2025
[24]

B. Li, H. Peng, R. Sainju, J. Yang, L. Yang, Y. Liang, W. Jiang, B. Wang, H. Liu, and C. Ding. Detecting Gender Bias in Transformer-based Models: A Case Study on BERT.arXiv preprint arXiv:2110.15733, 2021

2021 arXiv
[25]

McInnes, J

L. McInnes, J. Healy, and J. Melville. UMAP: Uniform Manifold Approximation and Projec- tion for Dimension Reduction.arXiv preprint arXiv:1802.03426, 2020

2020 arXiv
[26]

McInnes, J

L. McInnes, J. Healy, N. Saul, and L. Grossberger. UMAP: Uniform Manifold Approximation and Projection.The Journal of Open Source Software, 3(29):861, 2018

2018
[27]

A. S. Morcos, M. Raghu, and S. Bengio. Insights on representational similarity in neural networks with canonical correlation.arXiv preprint arXiv:1806.05759, 2018

2018 arXiv
[28]

Mordvintsev, C

A. Mordvintsev, C. Olah, and M. Tyka. Inceptionism: Going deeper into neural networks. Google Research Blog. Retrieved June, 20(14):5, 2015

2015
[29]

Nagamine, M

T. Nagamine, M. L. Seltzer, and N. Mesgarani. Exploring how deep neural networks form phonemic categories. InConference of the International Speech Communication Association (Interspeech), 2015

2015
[30]

Osindero and G

S. Osindero and G. E. Hinton. Modeling image patches with a directed hierarchy of Markov random fields. InAdvances in Neural Information Processing Systems, pages 1121–1128, 2008

2008
[31]

H. Park, N. Das, R. Duggal, A. P. Wright, O. Shaikh, F. Hohman, and D. H. P. Chau. Neu- rocartography: Scalable automatic visual summarization of concepts in deep neural networks. IEEE Transactions on Visualization and Computer Graphics, 28(1):813–823, 2021

2021
[32]

Salman, S

H. Salman, S. Jain, A. Ilyas, L. Engstrom, E. Wong, and A. Madry. When does bias transfer in transfer learning?arXiv preprint arXiv:2207.02842, 2022. 20 V. Krug and S. Stober

2022 arXiv
[33]

R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra. Grad-CAM: Visual Explanations From Deep Networks via Gradient-Based Localization. InIEEE Inter- national Conference on Computer Vision (ICCV), pages 618–626, 2017

2017
[34]

Simonyan and A

K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recog- nition.arXiv preprint arXiv:1409.1556, 2014

2014 arXiv
[35]

Stricker, F

P. Stricker, F. Röhrbein, and A. Knoblauch. Weight Perturbation and Competitive Heb- bian Plasticity for Training Sparse Excitatory Neural Networks. In2024 International Joint Conference on Neural Networks (IJCNN), pages 1–8, 2024

2024
[36]

L. Sweeney. Discrimination in Online Ad Delivery.arXiv preprint arXiv:1301.6822, 2013

2013 arXiv
[37]

Szegedy, V

C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the inception archi- tecture for computer vision. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016

2016
[38]

Torrey and J

L. Torrey and J. Shavlik. Transfer learning. InHandbook of research on machine learning applications and trends: algorithms, methods, and techniques, pages 242–264. IGI global, 2010

2010
[39]

Yosinski, J

J. Yosinski, J. Clune, A. Nguyen, T. Fuchs, and H. Lipson. Understanding Neural Networks Through Deep Visualization.arXiv preprint arXiv:1506.06579, 2015

2015 arXiv
[2023]

CEUR-WS, 2023

Kraków, Poland. CEUR-WS, 2023

2023

Pith tools

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

[1] [1]

Ahn and A

J. Ahn and A. Oh. Mitigating language-dependent ethnic bias in BERT. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 533– 549, Online and Punta Cana, Dominican Republic, Nov. 2021. Association for Computational Linguistics

work page 2021

[2] [2]

Alain and Y

G. Alain and Y. Bengio. Understanding intermediate layers using linear classifier probes. In International Conference on Learning Representations (ICLR), Workshop Track Proceedings, 2017

work page 2017

[3] [3]

Baker, H

N. Baker, H. Lu, G. Erlikhman, and P. J. Kellman. Deep convolutional networks do not classify based on global object shape.PLoS computational biology, 14(12):e1006613, 2018

work page 2018

[4] [4]

Bolukbasi, K.-W

T. Bolukbasi, K.-W. Chang, J. Y. Zou, V. Saligrama, and A. T. Kalai. Man is to computer programmer as woman is to homemaker? Debiasing Word Embeddings. InAdvances in Neural Information Processing Systems, volume 29, pages 4349–4357, 2016

work page 2016

[5] [5]

Buolamwini and T

J. Buolamwini and T. Gebru. Gender Shades: Intersectional Accuracy Disparities in Commer- cial Gender Classification. InProceedings of the 1st Conference on Fairness, Accountability and Transparency, volume 81, pages 77–91. PMLR, 23–24 Feb 2018

work page 2018

[6] [6]

Carter, Z

S. Carter, Z. Armstrong, L. Schubert, I. Johnson, and C. Olah. Activation atlas.Distill, 4(3):e15, 2019

2019

[7] [7]

J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In2009 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 248–255. Ieee, 2009

work page 2009

[8] [8]

Devlin, M.-W

J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.arXiv preprint arXiv:1810.04805, 2019

arXiv 2019