Pith. sign in

REVIEW 5 major objections 5 minor 71 references

KG-DG claims that combining clinical lesion rules with a vision transformer via confidence fusion improves diabetic retinopathy generalization to unseen domains.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

KG-DG fuses YOLO-derived lesion features with a frozen ViT via confidence-based fusion and claims gains in diabetic retinopathy domain generalization, but the central KL-divergence mechanism and the MDG headline are contradicted by the paper's own tables.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Abstract and MDG claims don't survive contact with the paper's own tables, but the SDG results may hide a salvageable empirical core if provenance questions are answered. the 5 major comments →

arxiv 2509.02918 v1 pith:TRDYV6MA submitted 2025-09-03 cs.CV cs.AI

Single Domain Generalization in Diabetic Retinopathy: A Neuro-Symbolic Learning Approach

Midhat Urooj , Ayan Banerjee , Farhat Shaikh , Kuntal Thakur , Sandeep Gupta This is my paper
classification cs.CV cs.AI
keywords domain generalizationneuro-symbolic learningdiabetic retinopathylesion featuresvision transformersconfidence fusionfundus imagingout-of-distribution robustness
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 sets out to show that diabetic retinopathy grading can generalize to unseen hospital domains when a vision transformer is paired with a symbolic branch that reasons over clinical lesion features. The proposed KG-DG framework detects hemorrhages, exudates, and cotton-wool spots, encodes them into a structured feature vector, classifies them with a gradient-boosting model, and fuses the symbolic prediction with the transformer's softmax output using confidence-based rules. The authors report that this integration beats published single-domain and multi-domain generalization baselines, with up to 5.2% cross-domain accuracy gains, a 6% gain over a ViT baseline on APTOS, and a symbolic-only MDG average of 63.67%. If true, the result would mean that explicit clinical knowledge, not just more data or bigger models, can make medical image classifiers more portable across devices and populations, and that interpretable lesion features can act as a regularizer rather than a bottleneck.

Core claim

KG-DG's central claim is that the best way to generalize a DR classifier to an unseen fundus-image domain is not to make the neural network alone more robust, but to give it a parallel, expert-driven reasoning path. The symbolic path encodes clinical rules from Table 1 (e.g., neovascularization implies proliferative DR) into quantitative lesion features, learns a Gradient Boosting classifier on those features, and then combines the two branches by a confidence-weighted fusion rule. On four public datasets treated as four domains, the authors report that the fused model outperforms transformer and CNN baselines in most single-source settings and that the symbolic-only model reaches the best a

What carries the argument

The engine is a dual-branch classifier. One branch is an ImageNet-pretrained ViT (DeiT-Small, CvT-13, or T2T-ViT) whose classification head is trained with class-weighted cross-entropy. The other branch turns clinical rules into numbers: YOLOv11 detects lesion bounding boxes, a U-Net vessel segmenter optionally measures tortuosity, caliber, and branching angle, and these measurements form a structured feature vector F* that a Gradient Boosting classifier maps to a severity label. A fusion step, selecting the highest-confidence prediction, the class-wise max, or a weighted sum of softmax scores, produces the final output. The framework also states that KL divergence between symbolic domain em

Load-bearing premise

The load-bearing premise is that all four datasets can be treated as one five-class DR task, including Messidor's grading scale, and that the lesion detectors were fine-tuned on a small expert set that does not include the evaluation domain, so the symbolic features are genuinely domain-invariant.

What would settle it

Retrain the YOLOv11 lesion detector from scratch on a training domain with no test-domain images, then run the full KG-DG fusion on a held-out test domain: if the symbolic branch's cross-domain accuracy matches the reported 74.0% on APTOS when trained on Messidor, the claim survives; if it collapses, the symbolic features are not domain-invariant. Separately, add the KL-divergence loss to training and measure embedding distances; if removing it changes no accuracy, the alignment mechanism is not load-bearing.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the paper's claims hold, DR grading systems could be deployed to a new clinic with a single source dataset, because lesion-based reasoning carries over even when pixel statistics do not.
  • Clinical lesion counts such as exudate count and hemorrhage density become a cheap, interpretable regularizer; models that use them need not trade away accuracy for explainability.
  • The symbolic-only result in MDG suggests that, for DR, a small curated set of biomarkers can outperform large-image classifiers under domain shift, reducing the annotation load for new deployment sites.
  • The same dual-branch pattern could transfer to other medical imaging tasks where expert rules identify countable lesions or structures, such as OCT biomarkers or chest X-ray findings.

Where Pith is reading between the lines

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

  • The fusion logic is modular, so swapping YOLOv11 for a lighter detector or using lesion counts alone would isolate whether the gain comes from the clinical prior or from detector quality.
  • The vein-feature ablation suggests not all clinical biomarkers transfer; ranking symbolic features by their cross-domain embedding distance could identify which ones should be dropped at deployment time.
  • A natural test of the alignment claim is to insert an explicit KL-divergence loss on the symbolic domain embeddings and compare cross-domain accuracy with and without it.
  • The same two-branch design could be tried on OCT or chest X-ray tasks where countable lesions and expert rules exist, with the symbolic branch providing an auditable second opinion.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes KG-DG, a neuro-symbolic framework for diabetic retinopathy (DR) classification that combines a Vision Transformer (ViT) branch with a symbolic branch based on lesion detection (YOLOv11) and retinal vessel segmentation features. The symbolic features are classified by a Gradient Boosting model, and the two branches are fused via confidence-based strategies. The authors claim that the framework addresses both single-domain generalization (SDG) and multi-domain generalization (MDG) by minimizing KL divergence between domain embeddings, and report accuracy gains up to 5.2% over baselines on APTOS, EyePACS, Messidor-1, and Messidor-2, with a symbolic-only model achieving 63.67% average accuracy in MDG. The paper includes SDG tables per source domain, an MDG comparison table, and an ablation study on lesion versus vessel features.

Significance. If the central claims were supported, the paper would make a useful contribution: it would demonstrate that structured clinical lesion features, when combined with deep representations, improve cross-domain generalization in DR classification—a practically important problem. The idea of using knowledge-guided symbolic features as regularizers is plausible and aligned with prior work on concept bottleneck models and knowledge-guided medical imaging. However, the significance is undermined by the fact that the main proposed mechanism (KL divergence alignment of domain embeddings) is never actually defined or implemented in the methodology, and the headline MDG result is contradicted by the paper's own table. The experimental comparisons also suffer from unresolved questions about feature-extractor provenance and label remapping, making the empirical claims unverifiable as stated.

major comments (5)
  1. [Abstract, Sec. 3, Sec. 5.2] The abstract and related work claim that KG-DG 'addresses both SDG and MDG by minimizing the KL divergence between domain embeddings, thereby enforcing alignment of high-level clinical semantics.' However, no KL-divergence loss, domain-embedding definition, or alignment training step appears anywhere in the Methodology (Sec. 3). The only occurrence of 'KL' in the experimental sections is as an abbreviation for the knowledge-based classifier (e.g., 'Knowledge (KL)' in Tables 2-5 and 'ViT+KL' in Table 6). Thus the central alignment mechanism is absent; the paper conflates a classifier label with a domain-alignment objective.
  2. [Sec. 4.2, Table 6] The MDG claim is internally contradicted. Sec. 4.2 states that 'the knowledge-guided symbolic model (KL only) achieved the best average accuracy (63.67%),' but Table 6 lists ERM-ViT (CvT-13) and SPSD-ViT with averages of 65.5%, i.e., above 63.67%. Additionally, the fused ViT+KL method (Ours) achieves only 58.4% average accuracy, which is below both its own neural-only ViT (61.18%) and the KL-only model (63.67%), and below nearly all baselines. The abstract's assertion that 'the complete neuro-symbolic integration achieves the highest accuracy compared to existing published baselines' is therefore not supported by the reported data.
  3. [Sec. 3.2] The provenance of the fine-tuning images for the feature extractors is unspecified. The paper says that YOLOv11 and the vein segmentation module were fine-tuned on 'approximately 500' expert-annotated images but does not state which dataset those images came from. If any target-domain images (e.g., APTOS) were included, then the symbolic features for that target are partially in-domain at the feature-extractor level, and the cross-domain generalization claim—particularly the large APTOS gains in Tables 3 and 4—is contaminated. The authors must disclose the exact source of these fine-tuning images and confirm that no target-domain data were used in training the knowledge extractors.
  4. [Sec. 5.1] The paper treats APTOS, EyePACS, Messidor-1, and Messidor-2 as interchangeable classification tasks with the same label space, but it never states how the Messidor grading scale was remapped to the 5-class APTOS/EyePACS labels. Messidor uses a 0-3 severity scale (or 0-4 in some versions), while APTOS/EyePACS use 0-4. Without an explicit remapping procedure, the symbolic rules and the cross-domain accuracy numbers are not well defined. The authors should specify the label alignment and report per-class or per-dataset label distributions to confirm that the comparison is meaningful.
  5. [Sec. 3.1] Fusion hyperparameters are 'empirically tuned' with no stated validation procedure. The weighted fusion weights (α_DL, α_KL) and the selection among the four fusion strategies appear to be chosen based on the test domains, as no held-out validation set is described. This constitutes a form of test-set fitting for the reported 'best' results. The authors should specify how these choices were made, ideally with a validation split or a sensitivity analysis across the full range of weights.
minor comments (5)
  1. [Throughout] The notation 'KL' for the knowledge classifier is confusing because the same letters are used for Kullback-Leibler divergence in the text. Please rename the symbolic classifier (e.g., 'Sym' or 'KB') and reserve 'KL' for the divergence measure.
  2. [Sec. 3.2] The paper alternates between 'vein segmentation' and 'vessel segmentation'; retinal imaging typically refers to arteries and veins jointly as vessels. Please use consistent terminology.
  3. [Sec. 5.1] The text says 'MESSIDOR' while elsewhere 'Messidor' is used. Please unify capitalization and dataset name formatting across tables and text.
  4. [Sec. 2] Some references appear mismatched: e.g., [59] is cited as TransDR but the reference list entry [59] is a diffusion-based domain augmentation paper. Please verify all citation-to-reference mappings.
  5. [Table 6] The backbone columns for the proposed methods list parameter counts but not the specific ViT variant consistently (e.g., 'Vit (22M)' vs 'Vit(21.5M)'). Clarify which architecture was used for each row.

Circularity Check

1 steps flagged

No derivation reduces to its inputs by construction; the main issues are unsupported KL-claim, internal MDG contradiction, and unspecified extractor provenance, which are correctness/verifiability problems. One minor non-load-bearing self-citation is present.

specific steps
  1. other [Section 2.2, second paragraph (Related Work: The Role of Knowledge-Guided Systems)]
    "Similarly, DeepXSOZ [24] utilized symbolic EEG biomarkers, achieving a 38.5% performance gain over CNN baselines for seizure onset prediction."

    Reference [24] is authored by two members of this paper's team (Banerjee and Gupta), making it a self-citation. It is invoked as supporting evidence that symbolic biomarkers improve out-of-distribution robustness. However, the central KG-DG claim does not reduce to this citation: the paper's own held-out-domain experiments are the operative evidence, and the symbolic classifier's results are measured independently of the cited work. Thus this is a minor self-citation rather than a load-bearing circular step; it is flagged only because the scoring rubric treats one non-load-bearing self-citation as a mild issue.

full rationale

I examined the claimed derivation chain: extract symbolic lesion/vessel features with fine-tuned YOLOv11 and a vessel segmenter, train a Gradient Boosting classifier on source-domain features, fuse with a ViT via confidence rules, and evaluate on held-out domains. No equation in Section 3 makes the target-domain label a function of the source fit; the reported cross-domain accuracies are genuine held-out measurements rather than identities. The abstract's claim of minimizing KL divergence between domain embeddings is not implemented in Section 3, and the MDG discussion contradicts Table 6 (63.67% is called the best while 65.5% baselines are listed in the same paragraph, and ViT+KL is 58.4%, below both components and baselines). These are correctness errors, not reductions by construction. The fusion weights are described as 'empirically tuned' without a stated validation protocol, but the paper does not say they are fit to target-domain labels, so I cannot exhibit a statistical forcing. The fine-tuning provenance of the extractors (approximately 500 images of unspecified origin) is a serious verifiability risk: if target-domain images were included, the 'unseen domain' claim would be contaminated, but the paper does not state this, so it remains a conditional concern rather than a demonstrated circular step. The only self-citation is ref. [24], used as motivation for symbolic biomarkers; it is not load-bearing for the paper's own empirical results. Accordingly, there is no construction-level circularity, and the score reflects only the minor non-load-bearing self-citation.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

The symbolic branch's claims rest on clinical rules (Table 1), an assumed-compatible label space across four datasets with different grading protocols, and the transferability of detectors fine-tuned on an unspecified ~500-image annotation set. The only fitted numbers are the fusion weights, the class weights, and the per-source selection of a fusion strategy. No new physical entities are invented; the 'domain embeddings' whose KL divergence is minimized are asserted but never defined, which is treated as a claim-without-derivation rather than an entity.

free parameters (3)
  • Fusion weights (alpha_DL, alpha_KL) = not reported
    Weighted fusion in section 3.1 uses 'empirically tuned weights'; values and tuning protocol (which validation set) are not given, so the reported best accuracies partly select over these weights.
  • Fusion strategy selection = Non-weighted for APTOS; Weighted for MESSIDOR2; varies by source
    The paper reports the best of four fusion rules per source domain (Tables 2-5); strategy chosen per dataset acts as a fitted hyperparameter.
  • Class weights in cross-entropy loss = not reported
    Section 3.3 uses 'class-weighted cross-entropy loss' with unspecified weights; affects the ViT branch's predictions.
axioms (4)
  • domain assumption Clinical lesion rules in Table 1 correctly map lesion presence and counts to DR severity grades across all four datasets.
    The symbolic classifier's labels are derived from expert rules (section 3.2); if the mapping is wrong for a dataset's grading protocol, the 'domain-invariant' features are not invariant.
  • domain assumption APTOS, EyePACS, Messidor-1, and Messidor-2 share a compatible label space (severity grades) without remapping.
    Section 5.1 treats each dataset as a domain for the same classification task but never describes grade remapping, despite Messidor's different grading scale.
  • domain assumption YOLOv11 and the vessel segmentation U-Net, fine-tuned on about 500 expert-annotated images, generalize feature extraction to all four datasets.
    Section 3.2: the 500-image annotation set's source dataset and distribution are unspecified; the symbolic features for every target domain depend on this.
  • domain assumption Lesion features are domain-invariant while vessel features are not.
    Ablation II (section 5.3) shows adding vessel features hurts APTOS accuracy; the framework's reliance on lesion features assumes this transferability holds across the other domains.
invented entities (1)
  • Domain embeddings of high-level clinical semantics (KL-aligned) no independent evidence
    purpose: The abstract says minimizing their KL divergence enforces domain alignment; no construction or loss function is given in section 3.
    The construct appears in the abstract and section 2.2 but is never defined mathematically or implemented; no equation, no training term, no evaluation of alignment.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Single Domain Generalization in Diabetic Retinopathy: A Neuro-Symbolic Learning Approach." pith.science (2026). https://pith.science/paper/TRDYV6MA

@misc{pith2026250902918,
  author       = {Pith},
  title        = {Pith review of: Single Domain Generalization in Diabetic Retinopathy: A Neuro-Symbolic Learning Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TRDYV6MA}},
  note         = {Machine review of arXiv:2509.02918}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Domain generalization remains a critical challenge in medical imaging, where models trained on single sources often fail under real-world distribution shifts. We propose KG-DG, a neuro-symbolic framework for diabetic retinopathy (DR) classification that integrates vision transformers with expert-guided symbolic reasoning to enable robust generalization across unseen domains. Our approach leverages clinical lesion ontologies through structured, rule-based features and retinal vessel segmentation, fusing them with deep visual representations via a confidence-weighted integration strategy. The framework addresses both single-domain generalization (SDG) and multi-domain generalization (MDG) by minimizing the KL divergence between domain embeddings, thereby enforcing alignment of high-level clinical semantics. Extensive experiments across four public datasets (APTOS, EyePACS, Messidor-1, Messidor-2) demonstrate significant improvements: up to a 5.2% accuracy gain in cross-domain settings and a 6% improvement over baseline ViT models. Notably, our symbolic-only model achieves a 63.67% average accuracy in MDG, while the complete neuro-symbolic integration achieves the highest accuracy compared to existing published baselines and benchmarks in challenging SDG scenarios. Ablation studies reveal that lesion-based features (84.65% accuracy) substantially outperform purely neural approaches, confirming that symbolic components act as effective regularizers beyond merely enhancing interpretability. Our findings establish neuro-symbolic integration as a promising paradigm for building clinically robust, and domain-invariant medical AI systems.

discussion (0)

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

Reference graph

Works this paper leans on

71 extracted references · 66 canonical work pages · 3 internal anchors

  1. [1]

    Arjovsky, L

    M. Arjovsky, L. Bottou, I. Gulrajani, and D. Lopez-Paz. Invariant risk minimization. arXiv preprint, arXiv:1907.02893, 2019

    Pith/arXiv arXiv 1907

  2. [2]

    Aslani, V

    S. Aslani, V . Murino, M. Dayan, R. Tam, D. Sona, and G. Hamarneh. Scanner invariant multiple sclerosis lesion segmentation from mri. In Proceedings of the IEEE 17th International Symposium on Biomedical Imaging (ISBI), pages 781–785, 2020

    work page 2020

  3. [3]

    Atwany and M

    M. Atwany and M. Yaqub. Drgen: Domain generalization in diabetic retinopathy classification. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 635–644, 2022

    work page 2022

  4. [4]

    Balaji, S

    Y . Balaji, S. Sankaranarayanan, and R. Chellappa. Metareg: Towards domain generalization using meta-regularization. In Proceedings of the 32nd International Conference on Neural Information Processing Systems (NeurIPS), pages 1006–1016, 2018

    work page 2018

  5. [5]

    Blanchard, G

    G. Blanchard, G. Lee, and C. Scott. Generalizing from several related classification tasks to a new unlabeled sample. InProceedings of the 24th International Conference on Neural Information Processing Systems (NeurIPS), pages 2178–2186, 2011

    work page 2011

  6. [6]

    Cha et al

    J. Cha et al. Swad: Domain generalization by seeking flat minima. In Proceedings of the 35th International Conference on Neural Information Processing Systems (NeurIPS), pages 22405–22418, 2021

    work page 2021

  7. [7]

    Chen et al

    C. Chen et al. Improving the generalizability of convolutional neural network–based segmentation on cmr images. Frontiers in Cardiovascular Medicine, 7:105, 2020

    work page 2020

  8. [8]

    T. Chen, K. Han, and J. Wang. Fortifying medical image domain generalization via contrastive feature disentanglement. IEEE Transactions on Medical Imaging, 43(2):512–525, 2024

    work page 2024

  9. [9]

    Decencière et al

    E. Decencière et al. Feedback on a publicly distributed image database: The messidor database. Image Analysis and Stereology, 2014

    work page 2014

  10. [10]

    Dosovitskiy et al

    A. Dosovitskiy et al. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations (ICLR), 2021

    work page 2021

  11. [11]

    Q. Dou, D. C. Castro, K. Kamnitsas, and B. Glocker. Domain generalization via model-agnostic learning of semantic features. In Advances in Neural Information Processing Systems (NeurIPS) , volume 32, pages 579–589, 2019. arXiv:1901.10184

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  12. [12]

    Espinosa Zarlenga et al

    A. Espinosa Zarlenga et al. Concept embedding models: Toward interpretable and accurate concept-based learning. In Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS), 2022

    work page 2022

  13. [13]

    A. A. for Pediatric Ophthalmology and Strabismus. Proliferative diabetic retinopathy. https://aapos.org/glossary/proliferative-diabetic-retinopathy, 2023

    work page 2023

  14. [14]

    R. N. Frank. Diabetic retinopathy. New England Journal of Medicine, 350(1):48–58, 2004

    work page 2004

  15. [15]

    Ganin, E

    Y . Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V . Lempitsky. Domain-adversarial training of neural networks. In Journal of Machine Learning Research, volume 17, pages 2096–2030, 2016

    work page 2096

  16. [16]

    Ghifary, W

    M. Ghifary, W. B. Kleijn, M. Zhang, and D. Balduzzi. Domain generalization for object recognition with multi-task autoencoders. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 2551–2559, 2015

    work page 2015

  17. [17]

    E. R. Group. Grading diabetic retinopathy and estimating its progression. Ophthalmology, 98(5):786–806, 1991

    work page 1991

  18. [18]

    Gulrajani and D

    I. Gulrajani and D. Lopez-Paz. In search of lost domain generalization. In International Conference on Learning Representations (ICLR), 2021

    work page 2021

  19. [19]

    S. Han, J. Wang, J. Luo, and D. Liu. Neuro-symbolic generative model for medical report generation with prior knowledge. IEEE Transactions on Medical Imaging, 40(12):3436–3447, 2021

    work page 2021

  20. [20]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

    work page 2016

  21. [21]

    Content Significance Distribution of Sub-Text Blocks in Articles and Its Application to Article-Organization Assessment

    C. Jayanga, G. Kuruppu, and M. H. Khan. Generalizing to unseen domains in diabetic retinopathy classification. arXiv preprint, arXiv:2311.01673, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  22. [22]

    Y . Jia, J. Zhang, S. Shan, and X. Chen. Single-side domain generalization for face anti-spoofing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8481–8490, 2020

    work page 2020

  23. [23]

    Diabetic retinopathy detection

    Kaggle. Diabetic retinopathy detection. https://www.kaggle.com/c/ diabeticretinopathy-detection, 2015

    work page 2015

  24. [24]

    Kamboj, A

    P. Kamboj, A. Banerjee, and S. K. S. Gupta. Expert knowledge driven human-ai collaboration for medical imaging: A study on epileptic seizure onset zone identification. IEEE Journal of Biomedical and Health Informatics, 2023. in press

    work page 2023

  25. [25]

    Kauppi et al

    T. Kauppi et al. The aptos 2019 blindness detection dataset. Kaggle,

    work page 2019

  26. [26]

    Khandelwal, R

    A. Khandelwal, R. Siyal, Y . Xu, and A. Bhaskaran. Graphdr: Lesion ontology guided graph convolution for diabetic retinopathy classification. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2023

    work page 2023

  27. [27]

    P. W. Koh, T. Nguyen, Y . S. Tang, S. Mussmann, E. Pierson, B. Kim, and P. Liang. Concept bottleneck models. In Proceedings of the 37th International Conference on Machine Learning (ICML), 2020

    work page 2020

  28. [28]

    D. Li, Y . Yang, Y . Song, and T. M. Hospedales. Deeper, broader and artier domain generalization. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 5543–5551, 2017

    work page 2017

  29. [29]

    D. Li, Y . Yang, Y . Song, and T. M. Hospedales. Learning to generalize: Meta-learning for domain generalization. In Proceedings of the 32nd AAAI Conference on Artificial Intelligence (AAAI) , volume 32, pages 427–434, 2018

    work page 2018

  30. [30]

    H. Li, S. Pan, S. Wang, and A. C. Kot. Domain generalization with adversarial feature learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5400–5409, 2018

    work page 2018

  31. [31]

    H. Li, Y . Wang, R. Wan, S. Wang, T.-Q. Li, and A. C. Kot. Domain generalization for medical imaging classification with linear-dependency regularization. In Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 3118–3129, 2020

    work page 2020

  32. [32]

    Y . Li, X. Tian, M. Gong, Y . Liu, T. Liu, K. Zhang, and D. Tao. Deep domain generalization via conditional invariant adversarial networks. In European Conference on Computer Vision (ECCV), 2018

    work page 2018

  33. [33]

    Y . Li, Y . Yang, W. Zhou, and T. M. Hospedales. Feature-critic networks for heterogeneous domain generalization. In Proceedings of the 36th International Conference on Machine Learning (ICML) , pages 3915–3924, 2019

    work page 2019

  34. [34]

    Li et al

    Y . Li et al. Deep domain generalization via conditional invariant adversarial networks. In Proceedings of the European Conference on Computer Vision (ECCV), pages 647–663, 2018

    work page 2018

  35. [35]

    Q. Liu, Q. Dou, and P.-A. Heng. Shape-aware meta-learning for generalizing prostate mri segmentation to unseen domains. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 475–485, 2020

    work page 2020

  36. [36]

    Q. Liu, Q. Dou, L. Yu, and P.-A. Heng. Ms-net: Multi-site network for improving prostate segmentation with heterogeneous mri data. IEEE Transactions on Medical Imaging, 39(9):2713–2724, Sept. 2020

    work page 2020

  37. [37]

    Mahajan, S

    D. Mahajan, S. Tople, and A. Sharma. Domain generalization using causal matching. In International Conference on Machine Learning (ICML), pages 7313–7324, 2021

    work page 2021

  38. [38]

    Muandet, D

    K. Muandet, D. Balduzzi, and B. Schölkopf. Domain generalization via invariant feature representation. In Proceedings of the 30th International Conference on Machine Learning (ICML), pages I–10–I–18, 2013

    work page 2013

  39. [39]

    A. A. of Ophthalmology. Diabetic retinopathy preferred practice pattern, 2023. https://www.aao.org/preferred-practice-pattern/ diabetic-retinopathy-ppp, 2023

    work page 2023

  40. [40]

    Otálora, M

    S. Otálora, M. Atzori, V . Andrearczyk, A. Khan, and H. Müller. Staining invariant features for improving generalization of deep convolutional neural networks in computational pathology.Frontiers in Bioengineering and Biotechnology, 7:198, 2019

    work page 2019

  41. [41]

    Ozkan and X

    E. Ozkan and X. Boix. On the benefits of multi-domain training for medical image analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12456–12466, 2024

    work page 2024

  42. [42]

    Publishing

    S. Publishing. Diabetic retinopathy. https://www.ncbi.nlm.nih.gov/ books/NBK560805/, 2024

    work page 2024

  43. [43]

    A. Rame, C. Dancette, and M. Cord. Fishr: Invariant gradient variances for out-of-distribution generalization. In International Conference on Machine Learning (ICML), 2022

    work page 2022

  44. [44]

    Sagawa, P

    S. Sagawa, P. W. Koh, T. Hashimoto, and P. Liang. Distributionally robust neural networks for group shifts. In International Conference on Learning Representations (ICLR), 2020

    work page 2020

  45. [45]

    D. M. Shama, J. Jing, and A. Venkataraman. Deepsoz: A robust deep model for joint temporal and spatial seizure onset localization from multichannel eeg data. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 183–193, 2023

    work page 2023

  46. [46]

    Shankar, V

    S. Shankar, V . Piratla, S. Chakrabarti, S. Chaudhuri, P. Jyothi, and S. Sarawagi. Generalizing across domains via cross-gradient training. In International Conference on Learning Representations (ICLR), 2018

    work page 2018

  47. [47]

    R. Shao, X. Lan, J. Li, and P. C. Yuen. Multi-adversarial discriminative deep domain generalization for face presentation attack detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10015–10023, 2019

    work page 2019

  48. [48]

    Y . Shi, J. Seely, P. H. S. Torr, N. Siddharth, A. Hannun, N. Usunier, and G. Synnaeve. Gradient matching for domain generalization. arXiv preprint, arXiv:2104.09937, 2021

    Pith/arXiv arXiv 2021

  49. [49]

    U. V . Shukla and K. Tripathy. Diabetic retinopathy, 2025. Updated August 25, 2023, https://www.ncbi.nlm.nih.gov/books/NBK560805/

    work page 2025

  50. [50]

    Singh, K

    R. Singh, K. Ramasamy, C. Abraham, V . Gupta, and A. Gupta. Diabetic retinopathy: An update. Indian Journal of Ophthalmology , 56(3): 179–188, 2008. doi: 10.4103/0301-4738.41167. https://www.ncbi.nlm. nih.gov/pmc/articles/PMC2636123/

    work page arXiv 2008

  51. [51]

    Sultana, M

    M. Sultana, M. Naseer, S. Khan, and F. Khan. Self-distilled vision transformer for domain generalization. InAsian Conference on Computer Vision (ACCV), 2022

    work page 2022

  52. [52]

    Sun and K

    B. Sun and K. Saenko. Deep coral: Correlation alignment for deep domain adaptation. In European Conference on Computer Vision (ECCV), pages 443–450. Springer, 2016

    work page 2016

  53. [53]

    B. Sun, J. Feng, and K. Saenko. Deep coral: Correlation alignment for deep domain adaptation. In European Conference on Computer Vision (ECCV), pages 443–450, 2016

    work page 2016

  54. [54]

    Touvron, M

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. Jégou. Training data-efficient image transformers & distillation through attention. In Proceedings of the 38th International Conference on Machine Learning (ICML), 2021

    work page 2021

  55. [55]

    Touvron, M

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. Jégou. Training data-efficient image transformers & distillation through attention. In International Conference on Machine Learning (ICML), 2021

    work page 2021

  56. [56]

    Vapnik.The Nature of Statistical Learning Theory

    V . Vapnik.The Nature of Statistical Learning Theory. Springer Science & Business Media, 1999

    work page 1999

  57. [57]

    V olpi and V

    R. V olpi and V . Murino. Addressing model vulnerability to distributional shifts over image transformation sets. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 7979–7988, 2019

    work page 2019

  58. [58]

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao. Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. arXiv preprint, arXiv:2207.02696, 2022

    Pith/arXiv arXiv 2022

  59. [59]

    Y . Wang, Z. Zhang, and X. Li. Diffusion-based domain augmentation for robust medical image analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12345–12355, 2024

    work page 2024

  60. [60]

    C. P. Wilkinson, F. L. Ferris, R. E. Klein, P. P. Lee, C.-D. Agardh, M. Davis, and H.-P. Hammes. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology, 110(9):1677–1682, 2003

    work page 2003

  61. [61]

    H. Wu, B. Xiao, N. Codella, M. Liu, X. Dai, L. Yuan, and L. Zhang. Cvt: Introducing convolutions to vision transformers. arXiv preprint, arXiv:2103.15808, 2021

    Pith/arXiv arXiv 2021

  62. [62]

    S. Yan, H. Song, N. Li, L. Zou, and L. Ren. Improve unsupervised domain adaptation with mixup training. In arXiv preprint, volume arXiv:2001.00677, 2020

    Pith/arXiv arXiv 2001

  63. [63]

    Yanoff and J

    M. Yanoff and J. S. Duker. Ophthalmology. Elsevier, 5th edition, 2019

    work page 2019

  64. [65]

    L. Yuan, Y . Chen, T. Wang, W. Yu, Y . Shi, Z. Jiang, F. E. H. Tay, J. Feng, and S. Yan. Tokens-to-token vit: Training vision transformers from scratch on imagenet. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021

    work page 2021

  65. [66]

    Zhang et al

    L. Zhang et al. Generalizing deep learning for medical image segmentation to unseen domains via deep stacked transformation. IEEE Transactions on Medical Imaging, 39(7):2531–2540, 2020

    work page 2020

  66. [67]

    Zhang, H

    M. Zhang, H. Marklund, N. Dhawan, A. Gupta, S. Levine, and C. Finn. Adaptive risk minimization: Learning to adapt to domain shift. In Advances in Neural Information Processing Systems (NeurIPS), 2021

    work page 2021

  67. [68]

    L. Zhao, Z. Yu, and G. Chen. Ddr: A diverse dataset for diabetic retinopathy classification. In Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 234–245, 2022

    work page 2022

  68. [69]

    K. Zhou, Y . Yang, T. M. Hospedales, and T. Xiang. Deep domain-adversarial image generation for domain generalisation. In Proceedings of the 34th AAAI Conference on Artificial Intelligence (AAAI), pages 13025–13032, 2020

    work page 2020

  69. [70]

    K. Zhou, Y . Yang, T. M. Hospedales, and T. Xiang. Learning to generate novel domains for domain generalization. In Proceedings of the European Conference on Computer Vision (ECCV), pages 561–578, 2020

    work page 2020

  70. [71]

    K. Zhou, Y . Yang, Y . Qiao, and T. Xiang. Domain generalization with mixstyle. In International Conference on Learning Representations (ICLR), 2021. arXiv:2012.03641

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  71. [2019]

    https://www.kaggle.com/c/aptos2019-blindness-detection

    work page

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.