The reviewed record of science sign in
Pith

arxiv: 2605.28780 · v1 · pith:UOXG7DAY · submitted 2026-05-27 · cs.CV · cs.LG

Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositions

Thomas Vitry , Kieran Edgeworth , Stefan Wermter , Jae Hee Lee This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-29 12:45 UTCgrok-4.3pith:UOXG7DAYrecord.jsonopen to challenge →

classification cs.CV cs.LG
keywords bias identificationspurious correlationsconcept decompositiongradient probinglabel-free auditingvision classifiersdistribution shiftworst-group accuracy
0
0 comments X

The pith

A post-hoc method identifies spurious concepts in frozen vision models by ranking NMF decompositions via gradient interactions on misclassified examples using only class labels.

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

Vision classifiers often exploit spurious correlations that break under distribution shift. The paper presents a label-free approach that decomposes intermediate activations from an audit set into concepts via non-negative matrix factorization. It then ranks those concepts according to how their activations align with backpropagated gradients on false positives and false negatives. Suppressing the highest-ranked concepts at inference time raises worst-group accuracy without any model updates or extra labels.

Core claim

Bias concepts leave a detectable gradient trail: they tend to activate when correcting false negatives and to be suppressed when correcting false positives; a bias estimator built from these interactions surfaces decision-relevant spurious directions that need not match any annotated attribute, enabling both auditing and direct mitigation by concept suppression.

What carries the argument

The bias estimator that ranks NMF concept vectors by the sign and magnitude of their interaction with gradients on misclassified examples from a class-label-only audit set.

If this is right

  • Suppressing the top-ranked concepts raises worst-group accuracy by up to 17.9 points on Waterbirds and 10.4 points on CelebA with no retraining.
  • The ranked concepts recover the known spurious cue on Colored MNIST and Waterbirds.
  • On CelebA the surfaced directions only partially overlap the annotated gender attribute.
  • The same procedure supplies both an interpretable audit and an actionable debiasing step for already-deployed models.

Where Pith is reading between the lines

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

  • The same gradient-probe logic could be tested on non-vision modalities if activations admit stable non-negative decompositions.
  • If the audit set is drawn from the training distribution rather than a held-out shift, the recovered directions may capture training-time shortcuts instead of shift-sensitive biases.
  • Iteratively re-ranking after each suppression round might expose chained or higher-order spurious directions the single-pass estimator misses.

Load-bearing premise

Gradient interactions of the decomposed concepts on misclassified audit examples reliably mark spurious decision directions even without group or attribute labels.

What would settle it

An experiment on Waterbirds or CelebA in which the top-ranked concepts are suppressed yet worst-group accuracy does not rise or falls compared with the unedited model.

Figures

Figures reproduced from arXiv: 2605.28780 by Jae Hee Lee, Kieran Edgeworth, Stefan Wermter, Thomas Vitry.

Figure 1
Figure 1. Figure 1: Diagram of the bias identification method. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Waterbirds concept bias analysis. 4.1 Concept–Bias Alignment We first investigate whether the spurious attributes learned by the model appear as distinct concept directions in the non-negative concept decomposition. We hypothesize that if the model relies on a spurious attribute to predict class y, then at least one concept in Wy aligns with that attribute. To test this, we use datasets with known ground-t… view at source ↗
Figure 4
Figure 4. Figure 4: Empirical validation of the gradient-concept interaction on a ResNet [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top-activating patches of representative bias concepts identified by our [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: CMNIST: cosine similarity between the estimated color-bias direction [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: CMNIST: bias score against bias-alignment score. Blue crosses represent [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Top-activating patches of bias concepts for CMNIST with corresponding [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Top-activating patches of common and uncommon bias concepts for [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Top-activating patches of common and uncommon bias concepts for [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
read the original abstract

Vision classifiers can exploit spurious correlations, achieving high in-distribution accuracy yet failing under distribution shift. Existing approaches to bias mitigation and analysis often depend on curated datasets, spurious-attribute or group labels, or retraining, which may be infeasible once a model is deployed or the relevant bias is unknown. We present a bias-label-free, post-hoc method for identifying spurious concepts in frozen vision models, relying only on standard class labels from a held-out audit dataset. For each target class, we collect patches from inputs predicted as that class and apply non-negative matrix factorization to intermediate activations to obtain a bank of interpretable concept vectors. Candidate concepts are then ranked with a bias estimator derived from their interaction with backpropagated gradients on misclassified examples: bias concepts tend to get activated when correcting false negatives and suppressed when correcting false positives. On Colored MNIST and Waterbirds the method recovers concepts aligned with the known spurious cue, and on CelebA it surfaces decision-relevant directions that only partially coincide with the annotated gender attribute; suppressing the top-ranked concepts at inference time improves worst-group accuracy by up to 17.9 percentage points on Waterbirds and 10.4 on CelebA without any retraining or parameter updates. Our method identifies decision-relevant spurious directions that need not coincide with annotated ones, providing both an interpretable auditing tool and an actionable debiasing handle for frozen vision models. Code is available at https://github.com/vitryt/label-free-bias-identification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes a post-hoc, label-free method to identify spurious concepts in frozen vision classifiers using only class labels from a held-out audit set. For each target class, patches from predicted inputs undergo NMF on intermediate activations to yield concept vectors; these are ranked by a bias estimator based on gradient interactions with misclassified examples (activated on false-negative corrections, suppressed on false-positive corrections). The method recovers known spurious cues on Colored MNIST and Waterbirds, surfaces partially gender-aligned directions on CelebA, and shows that suppressing the top-ranked concepts at inference improves worst-group accuracy by up to 17.9pp on Waterbirds and 10.4pp on CelebA without retraining or parameter updates. Code is provided.

Significance. If the gradient-probe ranking on misclassifications reliably isolates decision-relevant spurious directions (rather than core features or noise) from class-label-only data, the approach would offer a practical auditing and debiasing tool for deployed models where bias or group labels are unavailable. The reported gains without retraining are notable for real-world applicability, and the public code is a clear strength supporting reproducibility.

major comments (2)
  1. [Method section] Method section (bias estimator): the central claim that ranking NMF concepts by gradient sign patterns on false-negative vs. false-positive corrections reliably surfaces spurious cues (rather than core features) rests on an unverified mapping; no controls or ablations are described that test whether highest-ranked concepts correspond to the known bias cue versus other sources of misclassification.
  2. [Experiments section] Experiments (Waterbirds and CelebA results): the reported worst-group accuracy gains of 17.9pp and 10.4pp after suppressing top-ranked concepts are load-bearing for the actionable debiasing claim, yet the manuscript provides no baseline comparison (e.g., random concept suppression or suppression of bottom-ranked concepts) to establish that the gradient-based ranking step is necessary or effective.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below with clarifications based on the existing experiments and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Method section] Method section (bias estimator): the central claim that ranking NMF concepts by gradient sign patterns on false-negative vs. false-positive corrections reliably surfaces spurious cues (rather than core features) rests on an unverified mapping; no controls or ablations are described that test whether highest-ranked concepts correspond to the known bias cue versus other sources of misclassification.

    Authors: The Colored MNIST experiment functions as a direct control for the claimed mapping: the spurious cue (background color) is known a priori, and the bias estimator ranks the corresponding NMF concept highest, as reported in Section 4.1. This shows that gradient sign patterns on misclassifications prioritize the bias cue over core features. We agree that additional ablations would further strengthen the claim and will add a comparison against random concept selection in the revision. revision: partial

  2. Referee: [Experiments section] Experiments (Waterbirds and CelebA results): the reported worst-group accuracy gains of 17.9pp and 10.4pp after suppressing top-ranked concepts are load-bearing for the actionable debiasing claim, yet the manuscript provides no baseline comparison (e.g., random concept suppression or suppression of bottom-ranked concepts) to establish that the gradient-based ranking step is necessary or effective.

    Authors: The gains are obtained specifically from the top-ranked concepts identified by the gradient estimator, and the alignment with known spurious cues on Waterbirds provides supporting evidence for the ranking's relevance. We agree that explicit baselines would better isolate the contribution of the ranking step and will add comparisons to random concept suppression and bottom-ranked concepts in the revised experiments. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The bias estimator is defined directly from gradient sign patterns on false-negative vs. false-positive corrections using only class labels from a held-out audit set; NMF concepts are ranked by this construction and then suppressed at inference. Results are measured on external benchmarks (Waterbirds, CelebA) with reported worst-group accuracy gains. No self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. The procedure is independent of the target spurious labels it seeks to surface.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The method rests on assumptions that NMF yields decision-relevant concepts and that gradient signals on errors indicate spuriousness; free parameters include the NMF rank and ranking threshold, which are not quantified in the abstract.

free parameters (2)
  • NMF concept rank K
    Number of basis vectors extracted from activations; choice affects which concepts are available for gradient ranking.
  • bias ranking threshold
    Cutoff used to select top concepts for suppression; not specified in abstract.
axioms (2)
  • domain assumption NMF on intermediate activations produces interpretable concept vectors aligned with model decisions
    Invoked when collecting patches and building the concept bank for each target class.
  • domain assumption Gradient activation/suppression patterns on false negatives vs. false positives distinguish spurious from core concepts
    Core premise of the bias estimator described in the abstract.

pith-pipeline@v0.9.1-grok · 5809 in / 1425 out tokens · 50228 ms · 2026-06-29T12:45:59.522123+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 1 canonical work pages

  1. [1]

    In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    Allgeuer, P., Ahrens, K., Wermter, S.: Unconstrained open vocabulary image classifi- cation: Zero-shot transfer from text to image via clip inversion. In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). pp. 8217–8228. IEEE (2025)

    2025

  2. [2]

    Information Fusion77, 261–295 (2022)

    Anders, C.J., Weber, L., Neumann, D., Samek, W., Müller, K.R., Lapuschkin, S.: Finding and removing Clever Hans: Using explanation methods to debug and improve deep models. Information Fusion77, 261–295 (2022)

    2022

  3. [3]

    In: Proceedings of the 41st International Conference on Machine Learning

    Arefin, M.R., Zhang, Y., Baratin, A., Locatello, F., Rish, I., Liu, D., Kawaguchi, K.: Unsupervised Concept Discovery Mitigates Spurious Correlations. In: Proceedings of the 41st International Conference on Machine Learning. pp. 1672–1688. PMLR (2024)

    2024

  4. [4]

    Advances in Neural Information Processing Systems35, 23284–23296 (2022)

    Asgari,S.,Khani,A.,Khani,F.,Gholami,A.,Tran,L.,MahdaviAmiri,A.,Hamarneh, G.: MaskTune: Mitigating Spurious Correlations by Forcing to Explore. Advances in Neural Information Processing Systems35, 23284–23296 (2022)

    2022

  5. [5]

    In: Proceedings of the 37th International Conference on Machine Learning

    Bahng,H.,Chun,S.,Yun,S.,Choo,J.,Oh,S.J.: Learning De-biased Representations with Biased Representations. In: Proceedings of the 37th International Conference on Machine Learning. pp. 528–539. PMLR (2020)

    2020

  6. [6]

    Advances in Neural Information Processing Systems36, 66044–66063 (2023)

    Belrose, N., Schneider-Joseph, D., Ravfogel, S., Cotterell, R., Raff, E., Biderman, S.: LEACE: Perfect linear concept erasure in closed form. Advances in Neural Information Processing Systems36, 66044–66063 (2023)

    2023

  7. [7]

    In: The Thirty-ninth Annual Conference on Neural Information Processing Systems (2025)

    Bhusal, D., Clifford, M., Rampazzi, S., Rastogi, N.: FACE: Faithful Automatic Concept Extraction. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems (2025)

    2025

  8. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Chakraborty, R., Sletten, A., Kampffmeyer, M.C.: ExMap: Leveraging Explain- ability Heatmaps for Unsupervised Group Robustness to Spurious Correlations. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12017–12026 (2024)

    2024

  9. [9]

    Journal of Biomedical Informatics149, 104548 (2024)

    Correa, R., Pahwa, K., Patel, B., Vachon, C.M., Gichoya, J.W., Banerjee, I.: Efficient adversarial debiasing with concept activation vector — Medical image case-studies. Journal of Biomedical Informatics149, 104548 (2024)

    2024

  10. [10]

    Proceedings of the AAAI Conference on Artificial Intelligence38(19), 21046–21054 (2024)

    Dreyer, M., Pahde, F., Anders, C.J., Samek, W., Lapuschkin, S.: From Hope to Safety: Unlearning Biases of Deep Models via Gradient Penalization in Latent Space. Proceedings of the AAAI Conference on Artificial Intelligence38(19), 21046–21054 (2024)

    2024

  11. [11]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Fel, T., Picard, A., Béthune, L., Boissin, T., Vigouroux, D., Colin, J., Cadène, R., Serre, T.: CRAFT: Concept Recursive Activation FacTorization for Explainability. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2711–2721 (2023)

    2023

  12. [12]

    Nature Machine Intelligence2(11), 665–673 (2020) 14 T

    Geirhos, R., Jacobsen, J.H., Michaelis, C., Zemel, R., Brendel, W., Bethge, M., Wichmann, F.A.: Shortcut learning in deep neural networks. Nature Machine Intelligence2(11), 665–673 (2020) 14 T. Vitry et al

    2020

  13. [13]

    Advances in Neural Information Processing Systems37, 62481–62502 (2024)

    Gerych,W.,Zhang,H.,Hamidieh,K.,Pan,E.,Sharma,M.,Hartvigsen,T.,Ghassemi, M.: BendVLM: Test-Time Debiasing of Vision-Language Embeddings. Advances in Neural Information Processing Systems37, 62481–62502 (2024)

    2024

  14. [14]

    In: Workshop on Spurious Correlation and Shortcut Learning: Foundations and Solutions (2025)

    Ghaznavi, M., Asadollahzadeh, H., Noohdani, F.H., Tabar, S.V., Hasani, H., Al- vanagh, T.A., Rohban, M.H., Baghshah, M.S.: Exploiting What Trained Models Learn for Making Them Robust to Spurious Correlations without Group Annota- tions. In: Workshop on Spurious Correlation and Shortcut Learning: Foundations and Solutions (2025)

    2025

  15. [15]

    In: Advances in Neural Information Processing Systems

    Ghorbani, A., Wexler, J., Zou, J.Y., Kim, B.: Towards Automatic Concept-based Explanations. In: Advances in Neural Information Processing Systems. vol. 32. Curran Associates, Inc. (2019)

    2019

  16. [16]

    In: Findings of the Association for Computational Linguistics: ACL 2025

    Ghosh, S., Syed, R., Wang, C., Choudhary, V., Li, B., Poynton, C.B., Visweswaran, S., Batmanghelich, K.: LADDER: Language-Driven Slice Discovery and Error Rectification in Vision Classifiers. In: Findings of the Association for Computational Linguistics: ACL 2025. pp. 22935–22970. Association for Computational Linguistics (2025)

    2025

  17. [17]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    He,K.,Zhang,X.,Ren,S.,Sun,J.: DeepResidualLearning forImage Recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 770–778 (2016)

    2016

  18. [18]

    In: The Thirteenth International Conference on Learning Representations (2024)

    He, Q., Xu, K., Yao, A.: EvA: Erasing Spurious Correlations with Activations. In: The Thirteenth International Conference on Learning Representations (2024)

    2024

  19. [19]

    In: The Thirteenth International Conference on Learning Representations (2024)

    Hirota, Y., Chen, M.H., Wang, C.Y., Nakashima, Y., Wang, Y.C.F., Hachiuma, R.: SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP. In: The Thirteenth International Conference on Learning Representations (2024)

    2024

  20. [20]

    In: Proceed- ings of the 41st International Conference on Machine Learning

    Holstege, F., Wouters, B., Giersbergen, N.V., Diks, C.: Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimation. In: Proceed- ings of the 41st International Conference on Machine Learning. pp. 18568–18610. PMLR (2024)

    2024

  21. [21]

    In: Proceedings of the 35th International Conference on Machine Learning

    Kim, B., Wattenberg, M., Gilmer, J., Cai, C., Wexler, J., Viegas, F., Sayres, R.: Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV). In: Proceedings of the 35th International Conference on Machine Learning. pp. 2668–2677. PMLR (2018)

    2018

  22. [22]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kim, B., Kim, H., Kim, K., Kim, S., Kim, J.: Learning Not to Learn: Training Deep Neural Networks With Biased Data. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9012–9020 (2019)

    2019

  23. [23]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kim, Y., Mo, S., Kim, M., Lee, K., Lee, J., Shin, J.: Discovering and Mitigating Visual Biases through Keyword Explanation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 11082–11092 (2024)

    2024

  24. [24]

    In: The Eleventh International Conference on Learning Representations (ICLR) (2023)

    Kirichenko, P., Izmailov, P., Wilson, A.G.: Last layer re-training is sufficient for robustness to spurious correlations. In: The Eleventh International Conference on Learning Representations (ICLR) (2023)

    2023

  25. [25]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kowal, M., Dave, A., Ambrus, R., Gaidon, A., Derpanis, K.G., Tokmakov, P.: Un- derstanding Video Transformers via Universal Concept Discovery. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10946–10956 (2024)

    2024

  26. [26]

    Advances in Neural Information Processing Systems37, 121598–121629 (2024)

    LaBonte, T., Hill, J.C., Zhang, X., Muthukumar, V., Kumar, A.: The Group Robust- ness is in the Details: Revisiting Finetuning under Spurious Correlations. Advances in Neural Information Processing Systems37, 121598–121629 (2024)

    2024

  27. [27]

    Transactions on Machine Learning Research (2024) Bias Identification via Gradient Probes on Concept Decompositions 15

    Le, P.Q., Schlötterer, J., Seifert, C.: Out of Spuriousity: Improving Robustness to Spurious Correlations without Group Annotations. Transactions on Machine Learning Research (2024) Bias Identification via Gradient Probes on Concept Decompositions 15

    2024

  28. [28]

    Lee, J.H., Mikriukov, G., Schwalbe, G., Wermter, S., Wolter, D.: Concept-Based Explanations in Computer Vision: Where Are We and Where Could We Go? In: Computer Vision – ECCV 2024 Workshops. pp. 266–287. Springer Nature Switzerland (2024)

    2024

  29. [29]

    arXiv preprint arXiv:2206.03680 (2023)

    Lee, J., Lee, J., Jung, S., Choo, J.: Improving Evaluation of Debiasing in Image Classification. arXiv preprint arXiv:2206.03680 (2023)

    work page arXiv 2023

  30. [30]

    In: Proceedings of the 38th International Conference on Machine Learning (ICML)

    Liu, E.Z., Haghgoo, B., Chen, A.S., Raghunathan, A., Koh, P.W., Sagawa, S., Liang, P., Finn, C.: Just train twice: Improving group robustness without training group information. In: Proceedings of the 38th International Conference on Machine Learning (ICML). pp. 6781–6792 (2021)

    2021

  31. [31]

    In: Proceedings of International Conference on Computer Vision (ICCV) (December 2015)

    Liu, Z., Luo, P., Wang, X., Tang, X.: Deep learning face attributes in the wild. In: Proceedings of International Conference on Computer Vision (ICCV) (December 2015)

    2015

  32. [32]

    In: The Thirteenth International Conference on Learning Representations (2024)

    Mulchandani, V., Kim, J.E.: Severing Spurious Correlations with Data Pruning. In: The Thirteenth International Conference on Learning Representations (2024)

    2024

  33. [33]

    In: Advances in Neural Information Processing Systems (NeurIPS)

    Nam, J., Cha, H., Ahn, S., Lee, J., Shin, J.: Learning from failure: De-biasing classifier from biased classifier. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 33, pp. 20673–20684 (2020)

    2020

  34. [34]

    In: Ethics and Fairness in Medical Imaging

    Olesen, V., Weng, N., Feragen, A., Petersen, E.: Slicing Through Bias: Explaining Performance Gaps in Medical Image Analysis Using Slice Discovery Methods. In: Ethics and Fairness in Medical Imaging. pp. 3–13. Springer Nature Switzerland (2025)

    2025

  35. [35]

    In: Interpretable AI: Past, Present and Future (2024)

    Paduraru,C.D.,Barbalau,A.,Filipescu,R.,Nicolicioiu,A.L.,Burceanu,E.: Concept- Drift: Uncovering Biases through the Lens of Foundation Models. In: Interpretable AI: Past, Present and Future (2024)

    2024

  36. [36]

    In: The Thirteenth International Conference on Learning Representations (2024)

    Pahde,F.,Dreyer,M.,Weckbecker,M.,Weber,L.,Anders,C.J.,Wiegand,T.,Samek, W.,Lapuschkin,S.: Navigating Neural Space: Revisiting Concept Activation Vectors to Overcome Directional Divergence. In: The Thirteenth International Conference on Learning Representations (2024)

    2024

  37. [37]

    Advances in Neural Information Processing Systems37, 105171–105199 (2024)

    Panousis, K.P., Ienco, D., Marcos, D.: Coarse-to-Fine Concept Bottleneck Models. Advances in Neural Information Processing Systems37, 105171–105199 (2024)

    2024

  38. [38]

    In: Proceedings of the 41st International Conference on Machine Learning

    Pezeshki, M., Bouchacourt, D., Ibrahim, M., Ballas, N., Vincent, P., Lopez-Paz, D.: Discovering Environments with XRM. In: Proceedings of the 41st International Conference on Machine Learning. pp. 40551–40569. PMLR (2024)

    2024

  39. [39]

    In: International Conference on Learning Representations (2019)

    Sagawa, S., Koh, P.W., Hashimoto, T.B., Liang, P.: Distributionally Robust Neural Networks. In: International Conference on Learning Representations (2019)

    2019

  40. [40]

    Advances in Neural Information Processing Systems36, 56553–56575 (2023)

    Tsirigotis,C.,Monteiro,J.,Rodriguez,P.,Vazquez,D.,Courville,A.C.: GroupRobust Classification Without Any Group Information. Advances in Neural Information Processing Systems36, 56553–56575 (2023)

    2023

  41. [41]

    Wah, C., Branson, S., Welinder, P., Perona, P., Belongie, S.: The Caltech-UCSD Birds-200-2011 Dataset. Tech. Rep. 2010-001, California Institute of Technology (2011)

    2011

  42. [42]

    In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

    Wu, H., Bezold, G., Gunther, M., Boult, T., King, M.C., Bowyer, K.W.: Consistency and Accuracy of CelebA Attribute Values . In: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). pp. 3258–3266. IEEE Computer Society, Los Alamitos, CA, USA (Jun 2023)

    2023

  43. [43]

    Transactions on Machine Learning Research (2025) 16 T

    Ye, W., Jiang, L., Xie, E., Zheng, G., Ma, Y., Cao, X., Guo, D., Qi, D., He, Z., Tian, Y., Porter, C.W., Coffee, M., Zeng, Z., Li, S., Wang, Z., Huang, T.H.K., Rehg, J.M., Kautz, H., Zhang, A.: The Clever Hans Mirage: A Comprehensive Survey on Spurious Correlations in Machine Learning. Transactions on Machine Learning Research (2025) 16 T. Vitry et al

    2025

  44. [44]

    Computer Sciences & Mathematics Forum9(1), 2 (2024)

    Zare, S., Nguyen, H.V.: Frustratingly Easy Environment Discovery for Invariant Learning. Computer Sciences & Mathematics Forum9(1), 2 (2024)

    2024

  45. [45]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhang,Z.,Feng,M.,Li,Z.,Xu,C.: Discover and Mitigate Multiple Biased Subgroups in Image Classifiers. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10906–10915 (2024)

    2024

  46. [46]

    bias color

    Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., Torralba, A.: Places: A 10 Million Image Database for Scene Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence40(6), 1452–1464 (2018) Bias Identification via Gradient Probes on Concept Decompositions 17 A Algorithm The bias identification procedure is summarized in Alg. 1. Algorithm ...

    2018