Pith. sign in

REVIEW 3 major objections 4 minor 44 references

Multi Attribute Bias Mitigation via Representation Learning

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read GMBM suppresses multiple known biases by first teaching them to the network, then pruning their directions from the backbone gradient.

desk verdict Two-stage multi-bias method with a reasonable new metric, but the 'provable invariance' claim is false as written and the bias-amplification evidence is mixed; still deserves referee time. read the letter →

arxiv 2509.03616 v1 pith:PPTPMKR7 submitted 2025-09-03 cs.CV

Rajeev Ranjan Dwivedi , Ankur Kumar , Vinod K Kurmi This is my paper · ORCID
classification cs.CV
keywords multi-biasmitigationspuriouscorrelationsshortcutlearninggradientsuppressionbiasamplificationmetricworst-groupaccuracyrepresentationfairness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that vision models can be made robust to several overlapping shortcuts at once, instead of the usual one-at-a-time fixes. It proposes GMBM, a two-stage training procedure: first, separate encoders are trained to recognize each known spurious attribute, and their features are attention-weighted into the main feature so the classifier must explicitly discount them; then the bias encoders are discarded and the backbone is fine-tuned with a penalty on any gradient component that points back along those bias directions. The result is a single compact network that, at test time, needs no bias labels or extra modules. The paper also argues that existing bias-amplification metrics break under subgroup imbalance and train-test distribution shift, and introduces SBA, a test-only weighted measure that isolates model-induced amplification.

What carries the argument

The load-bearing object is the orthogonal residual l_j from Eq. 1: the part of each bias encoder's penultimate feature that is perpendicular to the backbone's penultimate feature. Penalizing the squared dot product between the cross-entropy gradient and each l_j forces the backbone to be locally flat along every known bias direction. The attention-weighted fusion in ABIL is the supporting mechanism that makes these residuals meaningful: it trains bias encoders whose features align with the shortcuts the backbone actually uses, so the residuals isolate real spurious directions rather than arbitrary feature components.

What would settle it

Train GMBM on FB-CMNIST with the two labeled color biases, then construct a held-out split where a third unlabeled texture corruption conflicts with the digit label. If GMBM's accuracy on that hidden subgroup drops sharply while its SBA stays low, the claimed invariance covered only the known biases, not all spurious channels.

Watch

Extended reading notes

Core claim

At the paper's center is the claim that known spurious cues can be provably removed from a learned representation by a gradient penalty. In stage one (ABIL), each bias attribute j gets an encoder whose penultimate feature b_j is fused with the backbone feature h via softmax cosine attention, h' = h + sum alpha_j b_j, and the classifier trains on h'. In stage two, the residual l_j = b_j - ((h·b_j)/||h||^2) h is the bias component orthogonal to the task feature. The loss penalizes (∇_h L_ce · l_j)^2 for each j, suppressing gradient steps that steer the backbone back into a bias subspace. The paper calls this 'provable invariance to all known spurious channels' and reports that it preserves tas

Load-bearing premise

The debiasing guarantee depends on each bias encoder's penultimate feature being a complete and separate readout of that shortcut; if a bias encoder entangles the shortcut with task-relevant information, the orthogonal residual will not cleanly isolate the bias direction.

Editorial extensions

If this is right

  • At inference, GMBM uses only the debiased backbone and classifier; no bias labels, bias encoders, or architectural overhead remain.
  • Mitigating one shortcut no longer hands reliance to another, because every known bias direction is suppressed in the same gradient penalty.
  • Worst-group and bias-conflicting accuracy improve on both synthetic and real datasets, including under extreme bias ratios such as q = 0.99.
  • SBA provides a stable, test-only bias amplification measure that does not blow up when training and test distributions differ.

Reading between the lines

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

  • Stage 2 is a generic orthogonalization step: any linear feature direction, whether discovered by a label or by an unsupervised probe, could be fed into the same gradient-penalty mechanism.
  • SBA is logically separate from GMBM and could become a standard evaluation score for any debiasing method facing imbalanced subgroups or distribution shift.
  • If the bias encoders capture only part of a spurious signal, the orthogonal residual under-covers it and some bias will survive; the invariance guarantee is therefore conditional on the encoders' completeness.
  • A natural extension is to run stage 2 repeatedly as new biases are discovered after deployment, since the penalty is independent of how the bias direction was obtained.
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

3 major / 4 minor

Summary. The paper proposes GMBM, a two-stage framework for mitigating multiple simultaneous biases in image classification. Stage 1 (ABIL) trains per-attribute bias encoders and fuses their penultimate features with the backbone feature via softmax attention, forcing the classifier to see and discount spurious cues. Stage 2 (Gradient-Suppression Fine-Tuning) discards the bias encoders and fine-tunes the backbone with a penalty on the squared gradient component along each residual vector l_j, defined as the component of the bias feature orthogonal to the backbone feature. The authors claim this 'enforces provable invariance to all known spurious channels.' They also introduce a new metric, Scaled Bias Amplification (SBA), and evaluate on FB-CMNIST, CelebA, and a custom COCO split, reporting improved unbiased/bias-conflicting accuracy and reduced SBA compared to baselines.

Significance. If the central claims were correct, GMBM would be a valuable contribution to multi-bias mitigation: it requires only group labels during training, uses a single compact network at inference, and addresses the 'Whac-A-Mole' problem of interacting biases. The paper also attempts to address a real limitation of existing metrics by proposing SBA. However, the central theoretical claim of provable invariance is not mathematically supported, and the main empirical evidence for bias-amplification reduction relies on a newly introduced metric, while the established MABA metric shows the method can be worse than ERM. These issues undermine the paper's headline contributions.

major comments (3)
  1. [Section 3.2, Eq. (1)] The claim that the gradient penalty 'enforces provable invariance to all known spurious channels' is incorrect. Since l_j is defined as the component of b_j orthogonal to h, l_j contains no information about bias directions already aligned with h. The penalty sum (∇_h L · l_j)^2 therefore only constrains gradient components orthogonal to h; it cannot remove spurious signal already captured in h. For a concrete counterexample, let h = u + v with v the spurious direction and b_j = v. Then l_j is a mixture of u and v, and it is possible that ∇_h L · l_j = 0 while ∇_h L · v ≠ 0. The paper's own Section 6 admits that 'gender remains partly entangled with make-up cues despite mitigation,' which is inconsistent with the 'provable invariance' claim. This is a load-bearing error in the central derivation.
  2. [Table 6, q=0.99 row] The abstract claims GMBM 'halves multi-attribute bias amplification,' but on the established MABA metric the results contradict this. At q=0.99, GMBM's Base MABA mean is 26.66 vs. ERM's 14.54; Min-Support MABA is 26.64 vs. 14.69; Weighted MABA is 18.84 vs. 10.35. Thus GMBM amplifies bias more than ERM under the existing metric. The only metric supporting the claim is the newly introduced SBA (Table 9). Since SBA is introduced in the same paper and does not measure amplification relative to the training distribution, this is not sufficient evidence for the headline result.
  3. [Section 4.3, SBA definition] SBA is presented as a bias-amplification metric that 'disentangles model induced bias amplification from distributional differences,' but by construction it uses only test-set ground-truth counts. A model that perfectly matches the test label distribution would achieve SBA=0 even if it amplified training bias, while a model that faithfully preserves training bias will have positive SBA whenever train and test distributions differ. Thus SBA conflates distribution shift with model behavior rather than isolating model-induced amplification. This matters because SBA is the main evidence for the paper's central claim of bias-amplification reduction.
minor comments (4)
  1. [Section 4.1 vs Table 1] Implementation details state the CMNIST model was 'trained for 80 epochs, followed by 10 epochs of fine-tuning,' while Table 1 gives T1=6 and T2=3. The discrepancy needs clarification.
  2. [Table 3] The baseline names are inconsistent: 'Vanilla' in Table 3 becomes 'ERM' in Tables 6–9. Please unify terminology.
  3. [Section 3.2, Eq. (1)] The notation for the residual vector is not consistently defined: h_i vs H, b_j^i vs B_j. Please define all symbols and dimensions explicitly.
  4. [Table 6] For Min-Support MABA, the variance at q=0.90 for GMBM is 513.42, essentially identical to Base MABA (513.69). The authors should explain why the min-support variant does not reduce variance in this case.

Circularity Check

0 steps flagged · score 1.0 of 10

No load-bearing circularity; SBA is self-introduced but not equivalent to the training objective, and self-citations are peripheral.

full rationale

GMBM's derivation chain is not circular in the sense required by the analysis. ABIL and Gradient-Suppression Fine-Tuning are defined directly from the feature equations (Eq. 1 and the penalty in Sec. 3.2); the reported unbiased/bias-conflicting accuracies are standard external measures, and GMBM improves them independently of SBA. The self-citations (CosfairNet [8]; Kurmi et al. [17]) appear only in related-work lists and do not justify any load-bearing premise. The introduction of SBA is a metric proposal, not a fitted prediction: SBA is computed from test-set counts for all methods without optimizing GMBM against it, and no equation in the paper identifies SBA with the gradient penalty. The 'provable invariance' statement in Sec. 3.2 is mathematically under-supported because l_j is constructed to be orthogonal to h, so bias already parallel to h is invisible to the penalty; this is a rigor/correctness gap, not a reduction of the result to its inputs. The paper itself notes in Sec. 6 that gender remains partly entangled with makeup cues, further indicating the claim is stronger than what the construction establishes.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The method depends on hand-chosen hyperparameters (lambda_supp, beta) and on the assumption that bias encoders faithfully represent spurious subspaces. No new physical or architectural entities are introduced beyond the metric SBA, which is a scoring rule rather than an entity.

free parameters (5)
  • lambda_supp (gradient penalty weight) = 10^-2
    Hand-chosen; controls the strength of the bias suppression penalty in Stage 2, central to the method's behavior.
  • beta (bias loss weight) = 0.2
    Hand-chosen; balances the main classification loss and the bias classifier losses in ABIL.
  • Training epochs T1 and T2 = inconsistent: Table 1 says T1=6, T2=3; Section 4.1 says CMNIST 80+10, ResNet 20+10
    The paper gives contradictory training schedules, making the reported results hard to reproduce.
  • Min-support threshold tau = not specified
    Used for Min-Support MABA tables; without a value, the reported numbers cannot be reproduced.
  • Epsilon in SBA = not specified
    Used to avoid division by zero in the SBA weight; the specific value is not given.
assumptions (3)
  • domain assumption Bias attributes b_1, ..., b_k are known and labeled for all training samples.
    The method requires group labels for each spurious attribute during training (Section 3 and Algorithm 1).
  • domain assumption The penultimate features of the bias encoders capture the spurious signals in a form that is linearly separable and alignable with the backbone feature h_i.
    ABIL and GSFT rely on b_j^i accurately representing the bias subspace (Section 3.1, 3.2).
  • ad hoc to paper The gradient penalty term (sum over j of (nabla_h L_ce . l_j)^2) is sufficient to enforce invariance to all known biases.
    The paper claims 'provable invariance' (Section 3.2) but provides no formal proof; the penalty is a soft regularizer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi Attribute Bias Mitigation via Representation Learning." pith.science (2026). https://pith.science/paper/PPTPMKR7

@misc{pith2026250903616,
  author       = {Pith},
  title        = {Pith review of: Multi Attribute Bias Mitigation via Representation Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PPTPMKR7}},
  note         = {Machine review of arXiv:2509.03616}
}
read the original abstract

Real world images frequently exhibit multiple overlapping biases, including textures, watermarks, gendered makeup, scene object pairings, etc. These biases collectively impair the performance of modern vision models, undermining both their robustness and fairness. Addressing these biases individually proves inadequate, as mitigating one bias often permits or intensifies others. We tackle this multi bias problem with Generalized Multi Bias Mitigation (GMBM), a lean two stage framework that needs group labels only while training and minimizes bias at test time. First, Adaptive Bias Integrated Learning (ABIL) deliberately identifies the influence of known shortcuts by training encoders for each attribute and integrating them with the main backbone, compelling the classifier to explicitly recognize these biases. Then Gradient Suppression Fine Tuning prunes those very bias directions from the backbone's gradients, leaving a single compact network that ignores all the shortcuts it just learned to recognize. Moreover we find that existing bias metrics break under subgroup imbalance and train test distribution shifts, so we introduce Scaled Bias Amplification (SBA): a test time measure that disentangles model induced bias amplification from distributional differences. We validate GMBM on FB CMNIST, CelebA, and COCO, where we boost worst group accuracy, halve multi attribute bias amplification, and set a new low in SBA even as bias complexity and distribution shifts intensify, making GMBM the first practical, end to end multibias solution for visual recognition. Project page: http://visdomlab.github.io/GMBM/

Figures

Figures reproduced from arXiv: 2509.03616 by the authors.

Figure 1
Figure 1. Gradient suppression fine tuning To counter these issues, a range of bias-mitigation approaches have been developed. Most of these techniques operate under the assumption of a single, known spurious cue and often demand privi￾leged group annotations during training or validation [30, 27]. How￾ever, recent studies demonstrate that this simplification fails to cap￾ture the complexity of real-world datasets—such as Ima… view at source ↗
Figure 2
Figure 2. Overview of the Generalized Multi-Bias Mitigation (GMBM) Framework. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 40 canonical work pages

  1. [1]

    Bahng, S

    H. Bahng, S. Chun, S. Yun, J. Choo, and S. J. Oh. Learning de-biased representations with biased representations. InInternational conference on machine learning, pages 528–539. PMLR, 2020

    work page 2020

  2. [2]

    C. A. Barbano, B. Dufumier, E. Tartaglione, M. Grangetto, and P. Gori. Unbiased supervised contrastive learning. In International Conference on Learning Representations (ICLR), 2023

    work page 2023

  3. [3]

    Buolamwini and T

    J. Buolamwini and T. Gebru. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fair- ness, accountability and transparency, pages 77–91. PMLR, 2018

    work page 2018

  4. [4]

    Creager, J.-H

    E. Creager, J.-H. Jacobsen, and R. Zemel. Environment inference for invariant learning. In International Conference on Machine Learning , pages 2189–2200. PMLR, 2021

    work page 2021

  5. [5]

    Darlow, S

    L. Darlow, S. Jastrz˛ ebski, and A. Storkey. Latent adversarial debias- ing: Mitigating collider bias in deep neural networks. arXiv preprint arXiv:2011.11486, 2020

    arXiv 2011

  6. [6]

    A. J. DeGrave, J. D. Janizek, and S.-I. Lee. AI for radiographic COVID- 19 detection selects shortcuts over signal. Nature Machine Intelligence, 2021

    work page 2021

  7. [7]

    P. Dhar, J. Gleason, A. Roy, C. D. Castillo, and R. Chellappa. Pass: Protected attribute suppression system for mitigating bias in face recog- nition. In ICCV, pages 15087–15096, 2021

    work page 2021

  8. [8]

    R. R. Dwivedi, P. Kumari, and V . K. Kurmi. Cosfairnet:a parameter- space based approach for bias free learning. In 35th British Machine Vision Conference 2024, BMVC 2024, Glasgow, UK, November 25-28,

    work page 2024

Show all 44 references
  1. [9]

    Z. et al. Contrastive adapters for foundation model group robustness. NeurIPS, 2022

    2022

  2. [10]

    Eyuboglu, M

    S. Eyuboglu, M. Varma, K. Saab, J.-B. Delbrouck, C. Lee-Messer, J. Dunnmon, J. Zou, and C. Ré. Domino: Discovering systematic er- rors with cross-modal embeddings. ICLR, 2022

    2022

  3. [11]

    Geirhos, J.-H

    R. Geirhos, J.-H. Jacobsen, C. Michaelis, R. Zemel, W. Brendel, M. Bethge, and F. A. Wichmann. Shortcut learning in deep neural net- works. Nature Machine Intelligence, 2020

    2020

  4. [12]

    S. Gong, X. Liu, and A. K. Jain. Jointly de-biasing face recognition and demographic attribute estimation. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XXIX 16, pages 330–347. Springer, 2020

    2020

  5. [13]

    Hong and E

    Y . Hong and E. Yang. Unbiased classification through bias-contrastive and bias-balanced learning. Advances in Neural Information Processing Systems, 34:26449–26461, 2021

    2021

  6. [14]

    Hosseini and E

    E. Hosseini and E. Fedorenko. Large language models implicitly learn to straighten neural sentence trajectories to construct a predictive rep- resentation of natural language. Advances in Neural Information Pro- cessing Systems, 36:43918–43930, 2023

    2023

  7. [15]

    S. Jain, H. Lawrence, A. Moitra, and A. Madry. Distilling Model Fail- ures as Directions in Latent Space. In International Conference on Learning Representations, 2023

    2023

  8. [16]

    Kirichenko, P

    P. Kirichenko, P. Izmailov, and A. G. Wilson. Last layer re-training is sufficient for robustness to spurious correlations. ICLR, 2023

    2023

  9. [17]

    Kurmi, R

    V . Kurmi, R. Sharma, Y . Sharma, and V . P. Namboodiri. Gradient based activations for accurate bias-free learning. In AAAI Conference on Artificial Intelligence , 2022. URL https://api.semanticscholar.org/ CorpusID:247025906

    2022

  10. [18]

    O. Lang, Y . Gandelsman, M. Yarom, Y . Wald, G. Elidan, A. Hassidim, W. T. Freeman, P. Isola, A. Globerson, M. Irani, and I. Mosseri. Explain- ing in Style: Training a GAN to explain a classifier in StyleSpace. In The IEEE/CVF International Conference on Computer Vision (ICCV) , 2021

    2021

  11. [19]

    O. Lang, Y . Gandelsman, M. Yarom, Y . Wald, G. Elidan, A. Hassidim, W. T. Freeman, P. Isola, A. Globerson, M. Irani, et al. Explaining in style: Training a gan to explain a classifier in stylespace. InProceedings of the IEEE/CVF ICCV, pages 693–702, 2021

    2021

  12. [20]

    Li and C

    Z. Li and C. Xu. Discover the unknown biased attribute of an image classifier. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14970–14979, 2021

    2021

  13. [21]

    Z. Li, A. Hoogs, and C. Xu. Discover and mitigate unknown biases with debiasing alternate networks. In ECCV, pages 270–288. Springer, 2022

    2022

  14. [22]

    Z. Li, I. Evtimov, A. Gordo, C. Hazirbas, T. Hassner, C. C. Ferrer, C. Xu, and M. Ibrahim. A whac-a-mole dilemma: Shortcuts come in multiples where mitigating one amplifies others. In CVPR, pages 20071–20082, 2023

    2023

  15. [23]

    T.-Y . Lin, M. Maire, S. Belongie, L. Bourdev, R. Girshick, J. Hays, P. Perona, D. Ramanan, C. L. Zitnick, and P. Dollár. Microsoft coco: Common objects in context, 2015. URL https://arxiv.org/abs/1405. 0312

    2015

  16. [24]

    E. Z. Liu, B. Haghgoo, A. S. Chen, A. Raghunathan, P. W. Koh, S. Sagawa, P. Liang, and C. Finn. Just train twice: Improving group robustness without training group information. In ICML, pages 6781–

  17. [25]

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

    2015

  18. [26]

    J. Nam, H. Cha, S. Ahn, J. Lee, and J. Shin. Learning from failure: De-biasing classifier from biased classifier. In H. Larochelle, M. Ran- zato, R. Hadsell, M. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems, volume 33, pages 20673–20684. Cur- r...

    2020

  19. [27]

    J. Nam, H. Cha, S. Ahn, J. Lee, and J. Shin. Learning from Failure: De-biasing Classifier from Biased Classifier. In Advances in Neural Information Processing Systems, 2020

    2020

  20. [28]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al. Learning transferable visual models from natural language supervision. In International con- ference on machine learning, pages 8748–8763. PmLR, 2021

    2021

  21. [29]

    Raghu, J

    M. Raghu, J. Gilmer, J. Yosinski, and J. Sohl-Dickstein. Svcca: Singular vector canonical correlation analysis for deep learning dynamics and interpretability. Advances in neural information processing systems , 30, 2017

    2017

  22. [30]

    Sagawa*, P

    S. Sagawa*, P. W. Koh*, T. B. Hashimoto, and P. Liang. Distributionally robust neural networks. In International Conference on Learning Rep- resentations, 2020. URL https://openreview.net/forum?id=ryxGuJrFvS

    2020

  23. [31]

    Sarridis, C

    I. Sarridis, C. Koutlis, S. Papadopoulos, and C. Diou. Badd: Bias miti- gation through bias addition. arXiv preprint arXiv:2408.11439, 2024

    2024 arXiv

  24. [32]

    Sarridis, C

    I. Sarridis, C. Koutlis, S. Papadopoulos, and C. Diou. FLAC: Fairness- Aware Representation Learning by Suppressing Attribute-Class As- sociations . IEEE Transactions on Pattern Analysis & Machine In- telligence, 47(02):1148–1160, Feb. 2025. ISSN 1939-3539. doi: 10.1109/TPAMI....

    2025

  25. [33]

    Singla and S

    S. Singla and S. Feizi. Salient ImageNet: How to discover spurious features in Deep Learning? In International Conference on Learning Representations, 2022

    2022

  26. [34]

    E. A. Stanley, R. Souza, M. Wilms, and N. D. Forkert. Where, why, and how is bias learned in medical image analysis models? a study of bias encoding within convolutional networks using synthetic data. EBioMedicine, 111, 2025

    2025

  27. [35]

    Tartaglione, C

    E. Tartaglione, C. A. Barbano, and M. Grangetto. End: Entangling and disentangling deep representations for bias correction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 13508–13517, 2021

    2021

  28. [36]

    Torralba and A

    A. Torralba and A. A. Efros. Unbiased look at dataset bias. In CVPR 2011, pages 1521–1528. IEEE, 2011

    2011

  29. [37]

    Tsirigotis, J

    C. Tsirigotis, J. Monteiro, P. Rodriguez, D. Vazquez, and A. Courville. Group robust classification without any group information. NeurIPS, 2023

    2023

  30. [38]

    Zhang, N

    M. Zhang, N. S. Sohoni, H. R. Zhang, C. Finn, and C. Ré. Correct-n- contrast: A contrastive approach for improving robustness to spurious correlations. In ICML, 2022

    2022

  31. [39]

    D. Zhao, A. Wang, and O. Russakovsky. Understanding and evaluating racial biases in image captioning, 2021. URL https://arxiv.org/abs/2106. 08503

    2021

  32. [40]

    D. Zhao, J. Andrews, and A. Xiang. Men also do laundry: Multi- attribute bias amplification. In A. Krause, E. Brunskill, K. Cho, B. En- gelhardt, S. Sabato, and J. Scarlett, editors, Proceedings of the 40th In- ternational Conference on Machine Learning, volume 202 of Proceed-...

    2023

  33. [41]

    Z. Zhao, Y . Ziser, and S. B. Cohen. Layer by layer: Uncovering where multi-task learning happens in instruction-tuned large language mod- els. In Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, editors, Proceed- ings of the 2024 Conference on Empirical Methods in Natural Lan- gua...

    2024

  34. [42]

    Z. Zhu, W. Liang, and J. Zou. Gsclip: A framework for explaining dis- tribution shifts in natural language. arXiv preprint arXiv:2206.15007, 2022

    2022 arXiv

  35. [43]

    doi: 10.18653/v1/ 2024.emnlp-main.847

    Association for Computational Linguistics. doi: 10.18653/v1/ 2024.emnlp-main.847

    2024 doi

  36. [2024]

    URL https://papers.bmvc2024.org/0738.pdf

    BMV A, 2024. URL https://papers.bmvc2024.org/0738.pdf

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.