Pith. sign in

REVIEW 5 major objections 6 minor 30 references

Debiasing Classifiers by Amplifying Bias with Latent Diffusion and Large Language Models

T0 review · 5 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read DiffuBias generates bias-conflict images with a pretrained latent diffusion model and an LLM captioner, claiming state-of-the-art debiasing accuracy on BFFHQ, Dogs & Cats, and BAR.

desk verdict Novel diffusion + LLM debiasing pipeline with strong real-world results, but the load-bearing 'bias-conflict' extraction is unvalidated and the SOTA claim is overstated. read the letter →

arxiv 2411.16079 v1 pith:KZRLPDDW submitted 2024-11-25 cs.CV cs.AI

Donggeun Ko , Dongjun Lee , Namjun Park , Wonkyeong Shim , Jaekwang Kim This is my paper · ORCID
classification cs.CVcs.AI
keywords datasetbiasbias-conflictsampledebiasinglatentdiffusionmodeltext-to-imagegenerationlargelanguagecaptioninggeneralizedcrossentropyloss
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 aims to establish that dataset bias in image classifiers can be corrected without attribute labels, without training a generative model, and without human intervention in the generation phase. Its DiffuBias pipeline takes the hardest examples from a deliberately biased classifier, captions them with a large vision-language model, filters those captions, and feeds them to a pretrained latent diffusion model to synthesise new bias-conflict images. The authors report state-of-the-art accuracy on three real-world benchmarks—88.31% on BFFHQ, 94.33% on Dogs & Cats, and 85.23% on BAR at 5% bias-conflict ratio—while using less training time and lower CO2 emissions than GAN-based augmentation. They also report that the pipeline underperforms on low-resolution synthetic CIFAR-10, which they attribute to the captioner's difficulty describing noise corruptions and to resolution mismatch.

What carries the argument

The load-bearing mechanism is a four-stage pipeline. First, a classifier $f_B$ trained with the Generalized Cross Entropy loss $\mathcal{L}_{GCE}(p(x;\theta),y) = (1 - p_y(x;\theta)^q)/q$ is steered to amplify errors on bias-conflict samples, whose gradient gets weighted by $p_y(x;\theta)^q$. Second, the top-100 training images by cross-entropy loss are extracted as a proxy set $X_{b.c.}$ of conflict samples. Third, a frozen vision-language captioner (LLaVA-LLaMA-3-8b) produces three captions per extracted image, and a text filter $F_T$ keeps only captions containing the dataset's class-defining frequent words. Fourth, a pretrained latent diffusion model generates new images from the filtered captions, and the final debiased classifier $f_D$ is trained on the union $X_{b.c.} + X_{b.a.} + X_{generated}$. The whole argument hinges on the captioner's text capturing the bias-conflict attribute so that the diffusion model renders it into novel samples.

What would settle it

On a dataset with known bias attributes, count how many of the top-100 extracted samples are actually bias-conflict (for BAR at 1%, only 14 true conflict images exist). If most extracted samples are bias-aligned and the generated images fail to improve (or worsen) debiased test accuracy relative to training on the original data alone, the central claim would be refuted.

Watch

Extended reading notes

Core claim

DiffuBias is claimed to be the first debiasing method that generates bias-conflict samples with a pretrained stable diffusion model, requiring no training of any generative component. The paper's core claim is that a classifier deliberately trained with Generalized Cross Entropy loss reveals which training images conflict with the dataset's bias, and that text-to-image generation from captions of those images can amplify the conflict signal enough to shift a biased classifier to a debiased one. On the reported experiments this yields state-of-the-art test accuracy on BFFHQ (88.31%), Dogs & Cats (94.33%), and BAR (85.23%) at a 5% bias-conflict ratio, and also improves results at 1% ratios. A secondary claim is that the approach is substantially cheaper than GAN-based augmentation, needing about 3 hours versus 35 and 10 hours for the A2 and AmpliBias baselines in the BFFHQ comparison.

Load-bearing premise

The pipeline assumes that the top-100 highest-loss training images are predominantly genuine bias-conflict samples even when the true bias-conflict ratio is 0.5-1%, and that captions derived from mis-identified samples still yield useful conflict images rather than reinforcing the original bias.

Editorial extensions

If this is right

  • Dataset debiasing can be performed with only off-the-shelf pretrained models, removing the need for attribute labels and for training GANs or diffusion models from scratch.
  • If the reported numbers hold, DiffuBias sets the new best accuracy on BFFHQ, Dogs & Cats, and BAR at 1-5% bias-conflict ratios, surpassing both GAN-based and supervised baselines.
  • The text filter is essential to the method's success: ablating it drops accuracy sharply on BAR and Dogs & Cats (e.g., from 85.23% to 67.66% on BAR 5%).
  • The method's gains are concentrated in real-world high-resolution imagery; on synthetic CCIFAR-10 it falls below several baselines, so the paper's claim of general debiasing is qualified to domains the captioner and diffusion model can render.

Reading between the lines

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

  • The top-100 extraction is likely noisy at 0.5-1% conflict ratios (BAR at 1% has only 14 true conflict images), so the captions may partly describe bias-aligned content; a ground-truth overlap audit would measure how much noise the pipeline tolerates.
  • The reported carbon advantage depends on freezing the generative models; if deployment requires fine-tuning the diffusion model for new domains, as the authors propose for future work, the efficiency gap would narrow.
  • A testable extension would be to pair the captioner with a corruption-specific captioning step (e.g., naming the noise type) and to generate at the target resolution, which the authors suggest would address the CCIFAR-10 failure.
  • The same pipeline should transfer to other spurious-correlation tasks such as watermark detection or medical imaging only if a pretrained diffusion model can render the relevant bias attribute, which the authors acknowledge as a limitation.
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

5 major / 6 minor

Summary. The paper proposes DiffuBias, a pipeline for debiasing image classifiers by generating synthetic bias-conflict samples with a pretrained latent diffusion model. The method trains a deliberately biased classifier with GCE loss, extracts the top-K highest-loss training images as proxy bias-conflict samples, captions them with a vision-language model, filters the captions by class-relevant frequent words, generates new images from the filtered captions with Stable Diffusion, and retrains the classifier on the augmented dataset. Experiments are reported on CCIFAR-10, BFFHQ, Dogs & Cats, and BAR at several bias-conflict ratios, together with t-SNE and GradCAM analyses and a carbon-emission comparison against GAN-based baselines. The paper claims state-of-the-art performance on the real-world benchmarks and novelty as the first diffusion-based debiasing approach of this kind.

Significance. If the claims are substantiated, DiffuBias would be a useful contribution: it avoids training a generative model, reuses off-the-shelf captioning and diffusion components, and reports strong gains on BFFHQ, Dogs & Cats, and BAR, with a substantially cheaper generative pipeline than GAN-based alternatives. The paper also provides several qualitative analyses (t-SNE, GradCAM, generated-sample inspection) that are helpful for understanding the method. However, the state-of-the-art claim is only partially supported by the paper's own Table 3, the central assumption that top-K GCE-loss samples are bias-conflict samples is unvalidated at low conflict ratios, and key generative-model details are missing, which limits both the strength and reproducibility of the contribution.

major comments (5)
  1. [§3.3, Table 5] The load-bearing assumption that top-100 highest-GCE-loss samples are bias-conflict samples is not credible at the reported conflict ratios. Table 5 shows that BAR at 1.0% has only 14 true bias-conflict training images, BFFHQ at 0.5% has 96, and Dogs & Cats at 1.0% has 80, yet K=100 in all cases. Consequently, most extracted seeds cannot be genuine bias-conflict samples under the dataset's own annotations. Section 4.3.2 concedes that erroneous extraction occurs, but the paper never measures the precision of the top-K selection, never replaces top-K with the ground-truth conflict set in an ablation, and never quantifies how many generated images actually challenge the bias. Without this, the accuracy gains could be attributed to generic augmentation or label-preserving diversity rather than to generation of bias-conflict samples. This issue directly affects the abstract's claim that the method generates bias-conflict samples.
  2. [Table 3 and Abstract] The abstract and conclusion state that DiffuBias achieves state-of-the-art performance on benchmark datasets, but Table 3 contradicts this on CCIFAR-10: DiffuBias obtains 27.73/30.82/31.83/40.93 across 0.5%/1.0%/2.0%/5.0%, which is below AmpliBias (34.63/45.95/48.74/52.22), below LfF+BE at higher ratios, and below DisEnt at all ratios. Since the paper reports only means over three trials, without standard deviations, the claimed small margins on BFFHQ (e.g., 88.31 vs 87.34 for AmpliBias) are not shown to be statistically distinguishable. The current evidence supports a claim of competitive performance on some real-world datasets, not the stated state-of-the-art claim across benchmarks.
  3. [§3.4.1, §3.5, §4.1.2] The generative pipeline is not reproducible as specified. The paper refers to 'a pretrained latent diffusion model' and cites LDM, but does not state which checkpoint or version was used (e.g., Stable Diffusion v1.4/v1.5/v2), the prompt template for the LLaVA captioner, the number of generated images per prompt, the classifier-free guidance scale, the number of diffusion inference steps, the sampler, or the seed. It also does not explain how the 512x512-style generated images are resized to 32x32 for CCIFAR-10 beyond the sentence in §4.2.2. Given that the method's contribution is the generation pipeline, these omissions are material to verification and comparison.
  4. [Table 4] Table 4 is internally inconsistent with Table 3. The DiffuBias row reports Dogs & Cats 1.0% and 5.0% accuracies of 73.51 and 83.95, whereas Table 3 reports DiffuBias Dogs & Cats as 84.17 and 94.33. The Table 4 entries are exactly the AmpliBias values from Table 3, suggesting a copy error. Because Table 4 is the only ablation showing the effect of the text filter on Dogs & Cats, this inconsistency must be corrected and the ablation re-verified.
  5. [§4.3.2, §4.3.6] The paper's own qualitative and ablation analyses weaken the central narrative and need quantitative support. Section 4.3.2 shows generated BFFHQ images that are mislabeled or erroneously derived from bias-aligned extraction, yet the frequency of such failures is never measured. Section 4.3.6 reports that DiffuBias without the text filter outperforms DiffuBias with the filter on BFFHQ at 0.5% (78.94 vs 78.88) and 1.0% (80.67 vs 79.48), which is acknowledged in the text but contradicts the opening sentence of the same section. The authors should provide per-condition variance and a direct measurement of how often the pipeline produces genuine bias-conflict samples, rather than relying on selected examples.
minor comments (6)
  1. [Table 5] The BFFHQ 5.0% row is arithmetically inconsistent: 9,120 bias-align plus 960 bias-conflict equals 10,080, not the 19,200 images implied by the other rows, and 960 is 5% of 19,200, not of 10,080.
  2. [Eq. (3)] The extraction notation is imprecise: using argmax over the training set with a single argmax would return one sample, not a set of K samples, and the equation mixes CE loss with the GCE loss described in Eq. (1). The top-K operation should be written explicitly.
  3. [Fig. 3] The carbon-emission bar chart is hard to read because the numbers appear as an unlabelled sequence ('r 1,492 17 54,720 707 19,061') with no clear association to methods or axes.
  4. [Section 4.3.6] The sentence 'DiffuBias without text filter underperforms in BFFHQ, BAR and Dogs & Cats datasets' is immediately qualified by the opposite result on BFFHQ 0.5% and 1.0%; the wording should be revised to match the numbers.
  5. [Throughout] There are numerous typos and grammatical errors, including 'misleads correlations', 'forumlated', 'datgasests', 'modles', and 'Despite of no training cost'. A careful proofreading pass is needed.
  6. [References] Reference [30] is incomplete: the URL ends with '=' and the entry is missing the full link and access details.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DiffuBias is an empirical augmentation pipeline evaluated on independent test splits; the top-K heuristic and self-citations are not load-bearing reductions.

full rationale

The derivation chain is empirical rather than definitional: a biased classifier is trained with GCE loss (Eq. 1-2), the top-100 high-loss training samples are extracted (Section 3.3), captioned and filtered (Section 3.4), used to condition a pretrained latent diffusion model (Section 3.5), and the augmented training set is used to train a debiased classifier whose accuracy is measured on held-out test sets with dataset-defined bias-conflict labels (Tables 3 and 5). The top-K extraction is an assumption about which samples are bias-conflict, and Section 4.3.2 explicitly acknowledges that some extracted or generated samples are erroneous; however, an imperfect heuristic is a correctness or robustness concern, not a circular reduction. No parameter is fitted to the test labels and then repackaged as a prediction. The self-citations (A2 [1], AmpliBias [14]) appear only in related work and as baselines in Table 3; they are not used to justify the central pipeline or to import a uniqueness theorem. The GCE loss and benchmark protocols are cited from external sources ([20], [29]). Consequently, the SOTA claim is an empirical result contingent on the quality of the generation pipeline, not a conclusion forced by the method's definitions.

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

The central claim rests on empirical assumptions about pretrained models and the top-K loss heuristic rather than on a formal theory. The only tunable quantities are K, q, F, the number of generated images, and the undisclosed diffusion sampling parameters, all of which affect the final classifier accuracy.

free parameters (5)
  • K (number of top-loss samples extracted) = 100
    Set in Section 3.3, used to define the bias-conflict sample set X_b.c.; no tuning or justification beyond a conjecture.
  • q (GCE loss exponent) = 0.7
    Set in Section 4.1.2 following LfF [20] for fair comparison; still a free choice affecting which samples get high loss.
  • F (text filter top frequent words) = 2 * num_classes
    Set in Section 3.4.2, e.g., F=20 for CCIFAR10; hand-picked filter width.
  • Number of generated images = Not specified
    Section 4.2.4 says an equivalent number of bias-conflict samples to bias-aligned samples is generated, but the exact count per class is not reported; affects the training set composition.
  • Diffusion sampling configuration = Unspecified
    No guidance scale, number of denoising steps, or generation resolution is reported; these materially affect image quality and downstream accuracy.
assumptions (4)
  • domain assumption A classifier trained with GCE loss assigns higher loss to bias-conflict samples.
    Inherited from LfF [20]; the paper relies on this in Section 3.3 to select samples for captioning.
  • domain assumption The captioner (LLaVA-LLaMA-3-8b) produces captions that preserve both the class label and the bias-conflict attribute.
    Stated in Section 3.4.1; the text filter cannot recover attributes the captioner omits.
  • domain assumption The latent diffusion model generates images whose class and bias attributes match the caption semantics.
    Required in Section 3.5 for the generated set to act as valid bias-conflict training data.
  • domain assumption The text filter with top-F words removes noisy captions while keeping class-relevant ones.
    Used in Section 3.4.2; ablation shows it helps on BAR and Dogs & Cats but slightly hurts BFFHQ at low ratios.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Debiasing Classifiers by Amplifying Bias with Latent Diffusion and Large Language Models." pith.science (2026). https://pith.science/paper/KZRLPDDW

@misc{pith2026241116079,
  author       = {Pith},
  title        = {Pith review of: Debiasing Classifiers by Amplifying Bias with Latent Diffusion and Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KZRLPDDW}},
  note         = {Machine review of arXiv:2411.16079}
}
abstract

Neural networks struggle with image classification when biases are learned and misleads correlations, affecting their generalization and performance. Previous methods require attribute labels (e.g. background, color) or utilizes Generative Adversarial Networks (GANs) to mitigate biases. We introduce DiffuBias, a novel pipeline for text-to-image generation that enhances classifier robustness by generating bias-conflict samples, without requiring training during the generation phase. Utilizing pretrained diffusion and image captioning models, DiffuBias generates images that challenge the biases of classifiers, using the top-$K$ losses from a biased classifier ($f_B$) to create more representative data samples. This method not only debiases effectively but also boosts classifier generalization capabilities. To the best of our knowledge, DiffuBias is the first approach leveraging a stable diffusion model to generate bias-conflict samples in debiasing tasks. Our comprehensive experimental evaluations demonstrate that DiffuBias achieves state-of-the-art performance on benchmark datasets. We also conduct a comparative analysis of various generative models in terms of carbon emissions and energy consumption to highlight the significance of computational efficiency.

Figures

Figures reproduced from arXiv: 2411.16079 by the authors.

Figure 1
Figure 1. Our overall pipeline of DiffuBias. It has four main components: (a), (b), (c), and (d), each executed sequentially. a) First, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. t-SNE embeddings of BFFHQ using vanilla ResNet-18 and DiffuBias of label young and old. (a) demonstrates that [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparative analysis of generative models based on [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Generated synthetic bias-conflict samples from our proposed framework, DiffuBias. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Wrong images generated without text filter, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Examples of generated images using DiffuBias pipeline of BFFHQ dataset. The first row of images are samples [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: GradCAM results of 𝑓𝐵𝑖𝑎𝑠𝑒𝑑 and 𝑓𝐷𝑒𝑏𝑖𝑎𝑠𝑒𝑑 for three benchmark test dataset. The test set is often bias-conflict samples which are difficult for the model to predict since the vanilla model is biased. chin and peripheral facial areas, where signs of aging such as wrin￾kl…
Figure 8
Figure 8. Figure 8: Example images of biases present in benchmark dataset. From the top, the datasets are CCIFAR10, Dogs & Cats, BAR [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 14 canonical work pages

  1. [1]

    Jaeju An, Taejune Kim, Donggeun Ko, Sangyup Lee, and Simon S Woo. 2022. Aˆ 2: Adaptive Augmentation for Effectively Mitigating Dataset Bias. In Proceedings of the Asian Conference on Computer Vision . 4077–4092

    work page 2022

  2. [2]

    Hyojin Bahng, Sanghyuk Chun, Sangdoo Yun, Jaegul Choo, and Seong Joon Oh. 2020. Learning de-biased representations with biased representations. In International Conference on Machine Learning . PMLR, 528–539

    work page 2020

  3. [3]

    XTuner Contributors. 2023. XTuner: A Toolkit for Efficiently Fine-tuning LLM. https://github.com/InternLM/xtuner

    work page 2023

  4. [4]

    Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A Wichmann, and Wieland Brendel. 2018. ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness. arXiv preprint arXiv:1811.12231 (2018)

    arXiv 2018

  5. [5]

    Karan Goel, Albert Gu, Yixuan Li, and Christopher Ré. 2020. Model patching: Closing the subgroup performance gap with data augmentation. arXiv preprint arXiv:2008.06775 (2020)

    arXiv 2020

  6. [6]

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2020. Generative adversarial networks. Commun. ACM 63, 11 (2020), 139–144

    2020

  7. [7]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition . 770–778

    2016

  8. [8]

    Dan Hendrycks and Thomas Dietterich. 2019. Benchmarking neural net- work robustness to common corruptions and perturbations. arXiv preprint arXiv:1903.12261 (2019)

    arXiv 2019

Show all 30 references
  1. [9]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33 (2020), 6840–6851

    2020

  2. [10]

    Inwoo Hwang, Sangjun Lee, Yunhyeok Kwak, Seong Joon Oh, Damien Teney, Jin-Hwa Kim, and Byoung-Tak Zhang. 2022. SelecMix: Debiased Learning by Mixing up Contradicting Pairs. In ICML 2022: Workshop on Spurious Correlations, Invariance and Stability

    2022

  3. [11]

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. 2020. Analyzing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 8110–8119

    2020

  4. [12]

    Byungju Kim, Hyunwoo Kim, Kyungsu Kim, Sungjin Kim, and Junmo Kim. 2019. Learning not to learn: Training deep neural networks with biased data. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 9012–9020

    2019

  5. [13]

    Eungyeup Kim, Jihyeon Lee, and Jaegul Choo. 2021. Biaswap: Removing dataset bias with bias-tailored swapping augmentation. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 14992–15001

    2021

  6. [14]

    Donggeun Ko, Dongjun Lee, Namjun Park, Kyoungrae Noh, Hyeonjin Park, and Jaekwang Kim. 2023. AmpliBias: Mitigating Dataset Bias through Bias Amplification in Few-shot Learning for Generative Models. In Proceedings of the 32nd ACM International Conference on Information and Kno...

    2023

  7. [15]

    Jungsoo Lee, Eungyeup Kim, Juyoung Lee, Jihyeon Lee, and Jaegul Choo. 2021. Learning debiased representation via disentangled feature augmentation. Ad- vances in Neural Information Processing Systems 34 (2021), 25123–25133

    2021

  8. [16]

    Jungsoo Lee, Jeonghoon Park, Daeyoung Kim, Juyoung Lee, Edward Choi, and Jaegul Choo. 2022. BiasEnsemble: Revisiting the Importance of Amplifying Bias for Debiasing. arXiv preprint arXiv:2205.14594 (2022)

    2022 arXiv

  9. [17]

    Jongin Lim, Youngdong Kim, Byungjai Kim, Chanho Ahn, Jinwoo Shin, Eunho Yang, and Seungju Han. 2023. BiasAdv: Bias-Adversarial Augmentation for Model Debiasing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 3832–3841

    2023

  10. [18]

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2024. Visual instruc- tion tuning. Advances in neural information processing systems 36 (2024)

    2024

  11. [19]

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. 2015. Deep learning face attributes in the wild. In Proceedings of the IEEE international conference on computer vision. 3730–3738

    2015

  12. [20]

    Junhyun Nam, Hyuntak Cha, Sungsoo Ahn, Jaeho Lee, and Jinwoo Shin. 2020. Learning from failure: De-biasing classifier from biased classifier. Advances in Neural Information Processing Systems 33 (2020), 20673–20684

    2020

  13. [21]

    Utkarsh Ojha, Yijun Li, Jingwan Lu, Alexei A Efros, Yong Jae Lee, Eli Shechtman, and Richard Zhang. 2021. Few-shot image generation via cross-domain corre- spondence. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 10743–10752

    2021

  14. [22]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  15. [23]

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

    2022

  16. [24]

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. 2019. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. arXiv preprint arXiv:1911.08731 (2019)

    2019 arXiv

  17. [25]

    Victor Schmidt, Benoit Courty, and Amine Saboni. 2023. CodeCarbon: A Software Package for Tracking the Carbon Emissions of Machine Learning Computations. https://github.com/mlco2/codecarbon. GitHub repository

    2023

  18. [26]

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedan- tam, Devi Parikh, and Dhruv Batra. 2017. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE interna- tional conference on computer vision . 618–626

    2017

  19. [27]

    Enzo Tartaglione, Carlo Alberto Barbano, and Marco Grangetto. 2021. End: Entan- gling and disentangling deep representations for bias correction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition . 13508–13517

    2021

  20. [28]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 (2023)

    2023 arXiv

  21. [29]

    Zhilu Zhang and Mert Sabuncu. 2018. Generalized cross entropy loss for training deep neural networks with noisy labels.Advances in neural information processing systems 31 (2018)

    2018

  22. [30]

    Airplane-Snow

    Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xiang. 2021. Domain General- ization with MixStyle. In International Conference on Learning Representations . https://openreview.net/forum?id= SIG Proceedings Paper in LaTeX Format ACM, 2025 A Appendix A.1 Datasets We elaborate furt...

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.