Pith. sign in

REVIEW 5 major objections 5 minor 65 references

Robust AI-Generated Face Detection with Imbalanced Data

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A CLIP-based detector that combines worst-case loss reweighting with AUC ranking achieves the best accuracy and lowest error rate on the DFWild-Cup test set among the compared CLIP baselines.

desk verdict A plausible recombined loss for CLIP-based deepfake detection whose 0.004 AUC edge over their own baseline is within noise; needs error bars and a fix to the gamma inconsistency before the superiority claim can stand. read the letter →

arxiv 2505.02182 v1 pith:PKPPSJY5 submitted 2025-05-04 cs.CV

Yamini Sri Krubha , Aryana Hou , Braden Vester , Web Walker , Xin Wang , Li Lin , Shu Hu This is my paper · ORCID
classification cs.CV
keywords deepfakedetectionAI-generatedfaceCLIPclassimbalanceCVaRlossAUCsharpness-awareminimizationfeatureaugmentation
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 the usual weakness of deepfake detectors under class imbalance can be addressed by changing the training objective rather than resampling. On CLIP ViT-L/14 features, it trains a small three-layer classifier with a loss that reweights the hardest examples (CVaR over a vector-scaling loss) and adds an AUC ranking term, then flattens the loss landscape with sharpness-aware minimization. On the DFWild-Cup benchmark, this combination reports the highest AUC (0.969), accuracy (0.914), and F1, and the lowest equal error rate (0.093) among the two CLIP baselines. The claim matters because real-world deepfake data is heavily skewed toward real images, and detectors trained on balanced benchmarks often fail when deployment data is imbalanced.

What carries the argument

The load-bearing mechanism is the composite loss $L = L_{\mathrm{CVaR}} + \gamma L_{\mathrm{AUC}}$. The CVaR loss minimizes the average loss over the hardest $\alpha$-fraction of examples in each batch, using the hinge form $[\ell(\theta; F_i^*, Y_i^*) - \lambda]_+$, and inside it a vector-scaling loss applies different multiplicative and additive logit factors per class to correct imbalance. The AUC loss is a smoothed ranking loss that pushes real samples above fake samples, directly optimizing the AUC metric. Training uses sharpness-aware minimization to minimize the loss at a perturbed parameter point, seeking a flattened loss landscape for generalization. All of this sits on frozen CLIP ViT-L/14 features augmented with additive Gaussian noise before the classifier head.

What would settle it

Run the same training procedure on the DFWild-Cup training set with ten different random seeds and compute test AUC confidence intervals; if the intervals for the proposed model and Baseline 2 overlap, the claim of outperforming the baselines fails.

Watch

Extended reading notes

Core claim

The central discovery is that the CVaR-plus-AUC objective improves CLIP-based deepfake detection under imbalance, with the final model beating both baselines on every metric in Table II: AUC 0.969067 versus 0.964797 for the stronger baseline, accuracy 0.913737 versus 0.906901, and EER 0.093176 versus 0.096457. Ablations attribute the gain to the CVaR loss: replacing it with cross-entropy while keeping the AUC term drops AUC to 0.964737 and accuracy to 0.901367. The paper also reports that a three-layer MLP is optimal; deeper heads overfit.

Load-bearing premise

The reported edge over the stronger baseline is a difference of 0.004 in AUC, measured on one test split with one training seed and hyperparameters selected on the validation set; if that edge is noise, the central comparative claim collapses.

Editorial extensions

If this is right

  • On the DFWild-Cup test set, the proposed model beats both CLIP baselines on all six reported metrics (AUC 0.969067, accuracy 0.913737, F1 0.913693, precision 0.914162, recall 0.913617, EER 0.093176).
  • The ablation shows the CVaR term is the main source of improvement: replacing it with cross-entropy while keeping the AUC term drops AUC to 0.964737 and accuracy to 0.901367.
  • The architecture study shows that a three-layer MLP is the best head; deeper heads degrade, with AUC falling to 0.9446 at fifteen layers, indicating CLIP features already encode most of the discriminative signal.
  • The sensitivity analysis shows the model is stable across gamma values in the tested range, with F1 between 0.893 and 0.908, so the approach does not depend on a fragile loss-weight setting.

Reading between the lines

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

  • Beyond the paper's experiments, the same CVaR-plus-ranking objective should transfer to other frozen feature extractors, since nothing in the loss depends on CLIP specifically; testing it on a non-CLIP backbone would isolate the loss's contribution.
  • The paper evaluates only one benchmark and one split, so the robustness claim for unseen generators is extrapolated from CLIP's pretraining; evaluating on newly released diffusion models would be a direct test.
  • The requirement to know the empirical class distribution for the VS loss weights, which the paper itself notes as a limitation, means the method is not directly usable when class priors are unknown or change over time.
  • The reported gains come from a single-seed run, so multi-seed evaluation with confidence intervals is the natural next check before deploying the detector in practice.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes a deepfake face detection framework that combines frozen CLIP ViT-L/14 features with a trainable MLP, latent feature augmentation, a Conditional Value-at-Risk (CVaR) loss built on a vector-scaling (VS) loss, an AUC ranking loss, and sharpness-aware minimization (SAM). The method is evaluated on the DFWild-Cup benchmark against two CLIP-based baselines, and the authors report a test AUC of 0.969067, accuracy of 0.913737, and EER of 0.093176, concluding that their approach achieves state-of-the-art performance and superior generalization under imbalanced data. The code is released.

Significance. If the reported results were substantiated, the framework would be a useful integration of distributionally robust optimization and ranking-based losses for deepfake detection, and the public code release is a positive contribution. The method is clearly motivated and the paper includes ablations on loss composition and MLP depth. However, the empirical evidence as presented does not currently support the strong claims of statistical superiority and state-of-the-art performance: the key margin over Baseline 2 is small, the experiments use a single split and a single seed, and there are internal inconsistencies in the reported hyperparameters.

major comments (5)
  1. [Section IV-B, Table II] The central empirical claim is that the proposed method outperforms Baseline 2, with the key margin being AUC 0.969067 versus 0.964797, a difference of 0.00427, on a test set of only 1,548 real and 1,524 fake images. The paper reports no confidence intervals, no multiple seeds, and no paired significance test. At these sample sizes the standard error of AUC is roughly 0.003-0.004, so the observed gap is within sampling noise and the claim of 'significantly outperforming' is not established. The authors should report confidence intervals or error bars from multiple seeds and perform a paired test (e.g., DeLong's test for AUC) before drawing this conclusion.
  2. [Section IV-A4 and Section IV-C/IV-D] There is a direct contradiction about the optimal value of gamma. Section IV-A4 states that the grid search found alpha = 0.9 and gamma = 0.6 as optimal. However, Table III and Section IV-D use gamma = 0.9 for the proposed method, and Table V reports validation AUCs of 0.962341 for gamma = 0.6 and 0.955130 for gamma = 0.9, neither of which equals the reported test AUC of 0.969067. This makes it impossible to determine which configuration produced Table II and undermines the reproducibility of the central result. The authors must reconcile these numbers and specify which gamma corresponds to the test-set results.
  3. [Table I and Section I] The stated motivation is that deepfake datasets contain 'far fewer fake samples compared to authentic ones,' but the DFWild-Cup training split used in this paper has 219,470 fake samples and 42,690 real samples. The minority class in the actual experimental data is the real class, not the fake class. The paper should either explicitly acknowledge the direction of imbalance in the benchmark or reframe the motivation, because the current text misdescribes the data that the proposed loss functions are applied to.
  4. [Section IV-A3] Baseline 2 is not an independent published method; it is constructed from the authors' own framework in reference [15] by replacing the cross-entropy loss with the VS loss and substituting the AUC loss. The comparison therefore isolates the loss changes but does not constitute a comparison against an external state-of-the-art detector. Consequently, the conclusion's claim of 'state-of-the-art performance' is not supported by the two-baseline comparison set. The authors should either add independent published baselines or temper the claim to a loss-component analysis.
  5. [Section IV and Section V] The conclusion claims 'resilience to distribution shifts' and 'superior generalization,' but the experiments are conducted on a single test split of a single benchmark. No cross-dataset evaluation is presented, and the Future Work section explicitly defers evaluation on other datasets. Claims about distribution-shift robustness and generalization should be removed or supported by cross-dataset experiments, or at minimum the paper should state that such claims are not evaluated.
minor comments (5)
  1. [Section III-D, Eq. (2)] The description of the AUC loss should clarify how the surrogate logistic loss replaces the indicator function and whether m+ and m- are computed per mini-batch or over the full augmented dataset, since the normalization affects the gradient scale.
  2. [Section IV-B, Fig. 2a] The text describes the ROC curve as 'near-perfect with an AUC close to 1,' but the reported test AUC is 0.969. If the figure shows validation performance, this should be stated explicitly to avoid confusion.
  3. [Section III-D] The VS loss formula is difficult to parse because of missing parentheses and inconsistent subscript notation for the class-dependent scaling factors. Please write the loss in a cleaner form with explicit definitions of zeta_0, zeta_1, delta_0, and delta_1.
  4. [Table IV] The accuracy and F1 values for the 6-layer and deeper MLPs (e.g., accuracy 0.596 for 6 layers) are dramatically lower than their AUC values, which suggests training collapse or a thresholding artifact. The paper should report the evaluation protocol and explain these discrepancies.
  5. [Abstract and Footnote] The URL in the abstract is written with an underscore ('SP_CUP') while the text on page 1 writes it without ('SP CUP'). Please make the URL consistent and correct throughout.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical benchmark comparison, not a derivation from the method's own fitted constants.

full rationale

The paper's contribution is an empirical detection pipeline: frozen CLIP ViT-L/14 features are augmented and fed into a trainable MLP, optimized with a composite loss (CVaR with VS loss plus an AUC surrogate) and SAM. No analytical result is derived from the quantities being predicted; the reported AUC, accuracy, F1, and EER are measured on the held-out DFWild-Cup test set (Table II). Hyperparameters α and γ are tuned on a validation set, which is standard model selection and does not make the test-set numbers forced by construction. Baseline 2 is constructed from the authors' prior work [15], but it is used as a running baseline rather than as the proof of the central claim, and Baseline 1 [21] is external. The 0.004 AUC gap over Baseline 2, single-seed reporting, and the γ inconsistency between Section IV-A4 and Table III are statistical and reproducibility concerns, not evidence that the derivation reduces to its inputs. The limitation statement about needing the empirical class distribution concerns applicability, not circularity. The claim chain is therefore self-contained with respect to circularity, though its strength should be tempered by the lack of confidence intervals.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities. It recombines known losses and a pretrained model. The main unstated burden is the missing beta parameter and the hand-set VS loss hyperparameters, which the central result depends on but cannot be reproduced from the text.

free parameters (6)
  • alpha (CVaR quantile) = 0.9
    Controls the fraction of hardest examples used in the CVaR loss; tuned by grid search on the validation set (Section IV-A4).
  • gamma (AUC loss weight) = 0.6 in implementation details, 0.9 in Table III and Table V
    Trade-off between CVaR loss and AUC loss; the reported optimal value is inconsistent across sections, affecting reproducibility and the claimed result.
  • beta (feature augmentation scale) = not reported
    Appears in Eq. (F_i + beta*eta) as the scaling factor for additive noise, but no value is given in the implementation details; the result depends on this unstated parameter.
  • sigma (noise covariance scale) = sampled from U(0,1) per step
    The covariance of the additive noise is set to sigma^2 I with sigma drawn uniformly from 0 to 1; this is an ad hoc distribution choice with no stated justification.
  • VS loss hyperparameters (zeta_0, zeta_1, delta_0, delta_1) = zeta_0=1.2, zeta_1=0.8, delta_0=0.05, delta_1=-0.05 (delta_1 typo in text)
    Hand-set constants for the vector-scaling loss, chosen without a reported tuning procedure; the class weighting factors omega are not specified at all.
  • nu (SAM perturbation norm) = 0.1
    Controls the size of the perturbation in sharpness-aware minimization; set by hand with no sensitivity analysis.
assumptions (6)
  • domain assumption CLIP ViT-L/14 pretrained on 400 million image-text pairs provides transferable features that separate real from AI-generated faces.
    The entire method relies on frozen CLIP features being discriminative for deepfake detection; this is stated in Section III-B and is not independently validated in this paper.
  • domain assumption Adding zero-mean Gaussian noise to CLIP feature vectors preserves the class label and improves boundary robustness.
    The augmentation in Section III-C is a core component, but the paper provides no evidence that label preservation holds for the chosen sigma range.
  • standard math The logistic surrogate is a valid differentiable replacement for the indicator function in the AUC loss.
    This is a standard surrogate in ranking losses, stated in Section III-D without proof, which is acceptable as background math.
  • standard math CVaR minimization over the per-sample losses is a valid distributionally robust optimization objective.
    The CVaR formulation in Eq. (1) follows published DRO work cited in the paper; treated as a known result.
  • standard math The first-order Taylor approximation and dual-norm solution in Eq. (4) correctly approximate the sharpness-aware minimization objective.
    This is the standard SAM derivation from Foret et al., cited as [61]; assumed valid.
  • domain assumption The empirical class distribution of the training set is known and can be used to compute the VS loss weighting factors.
    The authors explicitly list this as a limitation in Section V: the weighting factor requires knowing the empirical class distribution, which may be impractical in real deployments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust AI-Generated Face Detection with Imbalanced Data." pith.science (2026). https://pith.science/paper/PKPPSJY5

@misc{pith2026250502182,
  author       = {Pith},
  title        = {Pith review of: Robust AI-Generated Face Detection with Imbalanced Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PKPPSJY5}},
  note         = {Machine review of arXiv:2505.02182}
}
read the original abstract

Deepfakes, created using advanced AI techniques such as Variational Autoencoder and Generative Adversarial Networks, have evolved from research and entertainment applications into tools for malicious activities, posing significant threats to digital trust. Current deepfake detection techniques have evolved from CNN-based methods focused on local artifacts to more advanced approaches using vision transformers and multimodal models like CLIP, which capture global anomalies and improve cross-domain generalization. Despite recent progress, state-of-the-art deepfake detectors still face major challenges in handling distribution shifts from emerging generative models and addressing severe class imbalance between authentic and fake samples in deepfake datasets, which limits their robustness and detection accuracy. To address these challenges, we propose a framework that combines dynamic loss reweighting and ranking-based optimization, which achieves superior generalization and performance under imbalanced dataset conditions. The code is available at https://github.com/Purdue-M2/SP_CUP.

Figures

Figures reproduced from arXiv: 2505.02182 by the authors.

Figure 1
Figure 1. Overview of our proposed model using CLIP ViT-L/14 for encoding input images and an MLP module trained with a robust loss function combining CVaR on visual similarity and weighted AUC. The optimization process leverages a flattened loss landscape to ensure robust deepfake detection between real and fake images. E. Optimization Last, to further improve the detector’s generalization ca￾pability, we optimize the detect… view at source ↗
Figure 2
Figure 2. Visualization of the ROC curve (a) and training dynamics (b) for our proposed method. TABLE III: Ablation study of loss function composition Method AUC Accuracy F1 Precision Recall CE + γAUC (γ = 0.9) 0.964737 0.901367 0.901338 0.901549 0.901287 CVaR + γAUC (γ = 0.9) (Ours) 0.969067 0.913737 0.913693 0.914162 0.913617 TABLE IV: Ablation study of MLP architecture variants MLP Layers AUC Accuracy F1 Precision Recall 3… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 33 canonical work pages

  1. [15]

    Robust clip-based detector for exposing diffusion model-generated images,

    L. L. Santosh, I. Amerini, X. Wang, and S. Hu, “Robust clip-based detector for exposing diffusion model-generated images,” arXiv. org , 2024

    work page 2024

  2. [1]

    Ai-face: A million-scale demographically annotated ai-generated face dataset and fairness benchmark,

    L. Lin, X. Wang, S. Hu, et al., “Ai-face: A million-scale demographically annotated ai-generated face dataset and fairness benchmark,” CVPR, 2025

    2025

  3. [2]

    A Self-Learning Multimodal Approach for Fake News Detection

    H. Chen, H. Guo, B. Hu, S. Hu, J. Hu, S. Lyu, X. Wu, and X. Wang, “A self-learning multimodal approach for fake news detection,” arXiv preprint arXiv:2412.05843, 2024

    work page Pith review arXiv 2024

  4. [3]

    Few-shot learning for misinformation detec- tion based on contrastive models,

    P. Zheng, H. Chen, S. Hu, B. Zhu, J. Hu, C.-S. Lin, X. Wu, S. Lyu, G. Huang, and X. Wang, “Few-shot learning for misinformation detec- tion based on contrastive models,” Electronics, vol. 13, no. 4, p. 799, 2024

    work page 2024

  5. [4]

    Detecting multimedia generated by large ai models: A survey,

    L. Lin, N. Gupta, Y . Zhang, H. Ren, C.-H. Liu, F. Ding, X. Wang, X. Li, L. Verdoliva, and S. Hu, “Detecting multimedia generated by large ai models: A survey,” arXiv preprint arXiv:2402.00045 , 2024

    arXiv 2024

  6. [5]

    Harnessing the power of text-image contrastive models for automatic detection of online misinformation,

    H. Chen, P. Zheng, X. Wang, S. Hu, B. Zhu, J. Hu, X. Wu, and S. Lyu, “Harnessing the power of text-image contrastive models for automatic detection of online misinformation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 923–932, 2023

    work page 2023

  7. [6]

    Open-eye: An open platform to study human performance on identifying ai-synthesized faces,

    H. Guo, S. Hu, X. Wang, M.-C. Chang, and S. Lyu, “Open-eye: An open platform to study human performance on identifying ai-synthesized faces,” in 2022 IEEE 5th International Conference on Multimedia Information Processing and Retrieval (MIPR), pp. 224–227, IEEE, 2022

    work page 2022

  8. [7]

    Vb-kgn: Variational bayesian kernel generation networks for motion image deblurring,

    Y . Fu, X. Zhu, X. Li, X. Wang, X. Wu, S. Hu, Y . Wu, S. Lyu, and W. Liu, “Vb-kgn: Variational bayesian kernel generation networks for motion image deblurring,” IEEE Transactions on Multimedia , 2024

    work page 2024

Show all 65 references
  1. [8]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” Advances in neural information processing systems , vol. 27, 2014

    2014

  2. [9]

    X- transfer: A transfer learning-based framework for robust gan-generated fake image detection,

    L. Zhang, H. Chen, S. Hu, B. Zhu, X. Wu, J. Hu, and X. Wang, “X- transfer: A transfer learning-based framework for robust gan-generated fake image detection,” arXiv preprint arXiv:2310.04639 , vol. 2, 2023

    2023 arXiv

  3. [10]

    Gan-generated faces detection: A survey and new perspectives,

    X. Wang, H. Guo, S. Hu, M.-C. Chang, and S. Lyu, “Gan-generated faces detection: A survey and new perspectives,” ECAI 2023, pp. 2533– 2542, 2023

    2023

  4. [11]

    Deep fakes: A looming challenge for privacy, democracy, and national security,

    B. Chesney and D. Citron, “Deep fakes: A looming challenge for privacy, democracy, and national security,”Calif. L. Rev., vol. 107, p. 1753, 2019

    2019

  5. [12]

    Deepfakes and disinformation: Ex- ploring the impact of synthetic political video on deception, uncer- tainty, and trust in news,

    C. Vaccari and A. Chadwick, “Deepfakes and disinformation: Ex- ploring the impact of synthetic political video on deception, uncer- tainty, and trust in news,” Social media+ society , vol. 6, no. 1, p. 2056305120903408, 2020

    2020

  6. [13]

    Deep learning for deepfakes creation and detection: A survey,

    T. T. Nguyen, Q. V . H. Nguyen, D. T. Nguyen, D. T. Nguyen, T. Huynh- The, S. Nahavandi, T. T. Nguyen, Q.-V . Pham, and C. M. Nguyen, “Deep learning for deepfakes creation and detection: A survey,” Computer Vision and Image Understanding , vol. 223, p. 103525, 2022

    2022

  7. [14]

    Exposing deepfake videos by detecting face warping artifacts,

    Y . Li and S. Lyu, “Exposing deepfake videos by detecting face warping artifacts,” arXiv preprint arXiv:1811.00656 , 2018

    2018 arXiv

  8. [16]

    Learning a deep dual-level network for robust deepfake detection,

    W. Pu, J. Hu, X. Wang, Y . Li, S. Hu, B. Zhu, R. Song, Q. Song, X. Wu, and S. Lyu, “Learning a deep dual-level network for robust deepfake detection,” Pattern Recognition, vol. 130, p. 108832, 2022

    2022

  9. [17]

    Smote: synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,” Journal of artificial intel- ligence research, vol. 16, pp. 321–357, 2002

    2002

  10. [18]

    Mesonet: a compact facial video forgery detection network,

    D. Afchar, V . Nozick, J. Yamagishi, and I. Echizen, “Mesonet: a compact facial video forgery detection network,” in 2018 IEEE international workshop on information forensics and security (WIFS) , pp. 1–7, IEEE, 2018

    2018

  11. [19]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, 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

  12. [20]

    Learning transferable visual models from natural language supervision,

    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 conference on machine learning , pp. 8748–8763, PMLR, 2021

    2021

  13. [21]

    Rais- ing the bar of ai-generated image detection with clip,

    D. Cozzolino, G. Poggi, R. Corvi, M. Nießner, and L. Verdoliva, “Rais- ing the bar of ai-generated image detection with clip,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4356–4366, 2024

    2024

  14. [22]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10684–10695, 2022

    2022

  15. [23]

    Diffusion-empowered autoprompt medsam,

    P. Huang, S. Hu, B. Peng, J. Zhang, H. Zhu, X. Wu, and X. Wang, “Diffusion-empowered autoprompt medsam,” arXiv preprint arXiv:2502.06817, 2025

    2025 arXiv

  16. [24]

    Masked conditional diffusion model for enhancing deepfake detection,

    T. Chen, S. Yang, S. Hu, Z. Fang, Y . Fu, X. Wu, and X. Wang, “Masked conditional diffusion model for enhancing deepfake detection,” arXiv preprint arXiv:2402.00541, 2024

    2024 arXiv

  17. [25]

    Robustly optimized deep feature decoupling network for fatty liver diseases detection,

    P. Huang, S. Hu, B. Peng, J. Zhang, X. Wu, and X. Wang, “Robustly optimized deep feature decoupling network for fatty liver diseases detection,” in International Conference on Medical Image Computing and Computer-Assisted Intervention , pp. 68–78, Springer, 2024

    2024

  18. [26]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” in Proceedings of the IEEE international conference on computer vision , pp. 2980–2988, 2017

    2017

  19. [27]

    Synthesizing black-box anti-forensics deepfakes with high visual quality,

    B. Fan, S. Hu, and F. Ding, “Synthesizing black-box anti-forensics deepfakes with high visual quality,” in ICASSP 2024-2024 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 4545–4549, IEEE, 2024

    2024

  20. [28]

    A style-based generator architecture for generative adversarial networks,

    T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 4401–4410, 2019

    2019

  21. [29]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems , vol. 33, pp. 6840– 6851, 2020

    2020

  22. [30]

    Eyes tell all: Irregular pupil shapes reveal gan-generated faces,

    H. Guo, S. Hu, X. Wang, M.-C. Chang, and S. Lyu, “Eyes tell all: Irregular pupil shapes reveal gan-generated faces,” in ICASSP 2022- 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2904–2908, IEEE, 2022

    2022

  23. [31]

    Exposing gan-generated faces using in- consistent corneal specular highlights,

    S. Hu, Y . Li, and S. Lyu, “Exposing gan-generated faces using in- consistent corneal specular highlights,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2500–2504, IEEE, 2021

    2021

  24. [32]

    Crossdf: Improving cross-domain deepfake detection with deep infor- mation decomposition,

    S. Yang, H. Guo, S. Hu, B. Zhu, Y . Fu, S. Lyu, X. Wu, and X. Wang, “Crossdf: Improving cross-domain deepfake detection with deep infor- mation decomposition,” arXiv preprint arXiv:2310.00359 , 2023

    2023 arXiv

  25. [33]

    Attacking identity semantics in deepfakes via deep feature fusion,

    B. Fan, Z. Jiang, S. Hu, and F. Ding, “Attacking identity semantics in deepfakes via deep feature fusion,” in 2023 IEEE 6th International Con- ference on Multimedia Information Processing and Retrieval (MIPR) , pp. 114–119, IEEE, 2023

    2023

  26. [34]

    Rich and poor texture con- trast: A simple yet effective approach for ai-generated image detection,

    N. Zhong, Y . Xu, Z. Qian, and X. Zhang, “Rich and poor texture con- trast: A simple yet effective approach for ai-generated image detection,” CoRR, 2023

    2023

  27. [35]

    Unmasking the artist: Discriminating human-drawn and ai- generated human face art through facial feature analysis,

    M.-Q. Nguyen, K.-D. Ho, H.-M. Nguyen, C.-M. Tu, M.-T. Tran, and T.-L. Do, “Unmasking the artist: Discriminating human-drawn and ai- generated human face art through facial feature analysis,” in 2023 In- ternational Conference on Multimedia Analysis and Pattern Recognition (MAP...

    2023

  28. [36]

    Detecting images generated by deep diffusion models using their local intrinsic dimensionality,

    P. Lorenz, R. L. Durall, and J. Keuper, “Detecting images generated by deep diffusion models using their local intrinsic dimensionality,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 448–459, 2023

    2023

  29. [37]

    Wavelet-packets for deepfake image analysis and detection,

    M. Wolter, F. Blanke, R. Heese, and J. Garcke, “Wavelet-packets for deepfake image analysis and detection,” Machine Learning , vol. 111, no. 11, pp. 4295–4327, 2022

    2022

  30. [38]

    Ai-generated image detection using a cross-attention enhanced dual-stream network,

    Z. Xi, W. Huang, K. Wei, W. Luo, and P. Zheng, “Ai-generated image detection using a cross-attention enhanced dual-stream network,” in 2023 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) , pp. 1463–1470, IEEE, 2023

    2023

  31. [39]

    Ausome: authenticating social media images using frequency analysis,

    N. Poredi, D. Nagothu, and Y . Chen, “Ausome: authenticating social media images using frequency analysis,” in Disruptive Technologies in Information Sciences VII , vol. 12542, pp. 44–56, SPIE, 2023

    2023

  32. [40]

    Synthbuster: Towards detection of diffusion model gener- ated images,

    Q. Bammey, “Synthbuster: Towards detection of diffusion model gener- ated images,” IEEE Open Journal of Signal Processing , vol. 5, pp. 1–9, 2023

    2023

  33. [41]

    Robust attentive deep neural network for detecting gan-generated faces,

    H. Guo, S. Hu, X. Wang, M.-C. Chang, and S. Lyu, “Robust attentive deep neural network for detecting gan-generated faces,” IEEE Access , vol. 10, pp. 32574–32583, 2022

    2022

  34. [42]

    Preserving fairness generalization in deepfake detection,

    L. Lin, X. He, Y . Ju, X. Wang, F. Ding, and S. Hu, “Preserving fairness generalization in deepfake detection,” in Proceedings of the IEEE/CVF 7 Conference on Computer Vision and Pattern Recognition , pp. 16815– 16825, 2024

    2024

  35. [43]

    Improving fairness in deepfake detection,

    Y . Ju, S. Hu, S. Jia, G. H. Chen, and S. Lyu, “Improving fairness in deepfake detection,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pp. 4655–4665, 2024

    2024

  36. [44]

    OpenCLIP,

    G. Ilharco, M. Wortsman, R. Wightman, C. Gordon, N. Carlini, R. Taori, A. Dave, V . Shankar, H. Namkoong, J. Miller, H. Hajishirzi, A. Farhadi, and L. Schmidt, “OpenCLIP,” July 2021

    2021

  37. [45]

    Robust light-weight facial affective behavior recognition with clip,

    L. Lin, S. Papabathini, X. Wang, and S. Hu, “Robust light-weight facial affective behavior recognition with clip,” MIPR, 2024

    2024

  38. [46]

    Robust light-weight facial affective behavior recognition with clip,

    L. Lin et al., “Robust light-weight facial affective behavior recognition with clip,” arXiv, 2024

    2024

  39. [47]

    Outlier robust adversarial training,

    S. Hu, Z. Yang, X. Wang, Y . Ying, and S. Lyu, “Outlier robust adversarial training,” in Asian Conference on Machine Learning , pp. 454–469, PMLR, 2024

    2024

  40. [48]

    Rank-based decomposable losses in machine learning: A survey,

    S. Hu, X. Wang, and S. Lyu, “Rank-based decomposable losses in machine learning: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  41. [49]

    Distributionally robust survival analysis: A novel fairness loss without demographics,

    S. Hu and G. H. Chen, “Distributionally robust survival analysis: A novel fairness loss without demographics,” in Machine Learning for Health , pp. 62–87, PMLR, 2022

    2022

  42. [50]

    Tkml-ap: Adversarial attacks to top- k multi-label learning,

    S. Hu, L. Ke, X. Wang, and S. Lyu, “Tkml-ap: Adversarial attacks to top- k multi-label learning,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 7649–7657, 2021

    2021

  43. [51]

    Sum of ranked range loss for supervised learning,

    S. Hu, Y . Ying, X. Wang, and S. Lyu, “Sum of ranked range loss for supervised learning,” Journal of Machine Learning Research , vol. 23, no. 112, pp. 1–44, 2022

    2022

  44. [52]

    Learning by minimizing the sum of ranked range,

    S. Hu, Y . Ying, X. Wang, and S. Lyu, “Learning by minimizing the sum of ranked range,” Advances in Neural Information Processing Systems , vol. 33, pp. 21013–21023, 2020

    2020

  45. [53]

    Meta-learning with heterogeneous tasks,

    Z. Si, S. Hu, K. Ji, and S. Lyu, “Meta-learning with heterogeneous tasks,” arXiv preprint arXiv:2410.18894 , 2024

    2024 arXiv

  46. [54]

    Robust clip-based detector for exposing diffusion model-generated images,

    L. Lin, I. Amerini, X. Wang, S. Hu, et al., “Robust clip-based detector for exposing diffusion model-generated images,” MIPR, 2024

    2024

  47. [55]

    Fairness in survival analysis with distributionally robust optimization,

    S. Hu and G. H. Chen, “Fairness in survival analysis with distributionally robust optimization,” Journal of Machine Learning Research , vol. 25, no. 246, pp. 1–85, 2024

    2024

  48. [56]

    Hu, Rank-Based Losses in Machine Learning and Deep Learning

    S. Hu, Rank-Based Losses in Machine Learning and Deep Learning . PhD thesis, State University of New York at Buffalo, 2022

    2022

  49. [57]

    Label- imbalanced and group-sensitive classification under overparameteriza- tion,

    G. R. Kini, O. Paraskevas, S. Oymak, and C. Thrampoulidis, “Label- imbalanced and group-sensitive classification under overparameteriza- tion,” Advances in Neural Information Processing Systems , vol. 34, pp. 18970–18983, 2021

    2021

  50. [58]

    Minimax auc fairness: Efficient algorithm with provable convergence,

    Z. Yang, Y . L. Ko, K. R. Varshney, and Y . Ying, “Minimax auc fairness: Efficient algorithm with provable convergence,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 37, pp. 11909–11917, 2023

    2023

  51. [59]

    Robust clip-based de- tector for exposing diffusion model-generated images,

    L. Lin, I. Amerini, X. Wang, S. Hu, et al. , “Robust clip-based de- tector for exposing diffusion model-generated images,” arXiv preprint arXiv:2404.12908, 2024

    2024 arXiv

  52. [60]

    Preserving auc fairness in learning with noisy protected groups,

    M. Wu, L. Lin, W. Zhang, X. Wang, Z. Yang, and S. Hu, “Preserving auc fairness in learning with noisy protected groups,” in The 42nd International Conference on Machine Learning (ICML) , 2025

    2025

  53. [61]

    Sharpness-aware minimization for efficiently improving generalization,

    P. Foret, A. Kleiner, H. Mobahi, and B. Neyshabur, “Sharpness-aware minimization for efficiently improving generalization,” arXiv preprint arXiv:2010.01412, 2020

    2010 arXiv

  54. [62]

    Improving generalization for ai-synthesized voice detection,

    H. Ren, L. Li, C.-H. Liu, X. Wang, and S. Hu, “Improving generalization for ai-synthesized voice detection,” AAAI, 2024

    2024

  55. [63]

    Deepfakebench: A comprehensive benchmark of deepfake detection,

    Z. Yan, Y . Zhang, X. Yuan, S. Lyu, and B. Wu, “Deepfakebench: A comprehensive benchmark of deepfake detection,” inAdvances in Neural Information Processing Systems (A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, eds.), vol. 36, pp. 4534–4565, Curran Asso...

    2023

  56. [64]

    Ieee sps signal processing cup at icassp 2025: Deepfake face detection in the wild,

    “Ieee sps signal processing cup at icassp 2025: Deepfake face detection in the wild,” in https://2025.ieeeicassp.org/sp-cup/

    2025

  57. [65]

    Deep fake detection and classification using error-level analysis and deep learning,

    R. Rafique, R. Gantassi, R. Amin, J. Frnda, A. Mustapha, and A. H. Al- shehri, “Deep fake detection and classification using error-level analysis and deep learning,” Scientific reports, vol. 13, no. 1, p. 7422, 2023. 8

    2023

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.