pith. sign in

arxiv: 2606.28134 · v1 · pith:62GAMQZWnew · submitted 2026-06-26 · 💻 cs.LG · cs.AI

Beyond Sparse Supervision: Diffusion-Guided Learning for Few-Shot Graph Fraud Detection

Liming Liu , Chao Hu , Mingfei Lu , Yiwei Ge , Xingle Li , Heyuan Shi This is my paper

Pith reviewed 2026-06-29 04:55 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords graph fraud detectionfew-shot learningdiffusion augmentationcontrastive learninggraph neural networksspectral attentionanomaly detectionsparse supervision
0
0 comments X

The pith

ADC-GNN uses diffusion to create noise-perturbed feature views and contrastive learning to stabilize them, yielding gains in graph fraud detection when only 1 percent of labels are available.

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

The paper introduces ADC-GNN to tackle sparse labels and representation dilution in fraud graphs, where message passing can oversmooth camouflaged anomalies. It treats diffusion as a feature-space augmentation that adds noise to node features on a cosine schedule and then pulls the resulting views together with contrastive loss. A spectral attention layer further weights multi-hop and relation signals that matter for fraud. Under the 1 percent training regime the method records consistent gains over both classic graph-fraud baselines and recent anomaly-detection models on public benchmarks, with supporting results on a private telecom dataset.

Core claim

ADC-GNN formulates diffusion as a feature-space denoising augmentation mechanism that constructs noise-perturbed node-feature views under a cosine schedule and uses contrastive learning to stabilize node representations across perturbations; the spectral attention module adaptively emphasizes fraud-relevant hop-level and relation-level cues, producing consistent improvements over original graph fraud baselines and four protocol-consistent recent graph anomaly/fraud baselines on public benchmarks when only 1 percent of training labels are supplied.

What carries the argument

Diffusion-guided feature augmentation that builds cosine-scheduled noise-perturbed node-feature views and stabilizes them through contrastive representation learning, paired with multi-hop spectral attention.

If this is right

  • Fraud detection models can maintain accuracy when verified labels drop to 1 percent of the data.
  • Feature-space diffusion augmentation combined with contrastive stabilization mitigates oversmoothing and frequency suppression that normally hide fraud signals.
  • Multi-hop spectral attention supplies relation-level and hop-level cues that standard spatial or spectral filters miss.
  • The same pipeline produces usable results on both public academic benchmarks and a private real-world telecom transaction graph of roughly 60,000 records.

Where Pith is reading between the lines

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

  • The same diffusion-contrastive recipe could be tested on other sparse-label graph tasks such as money-laundering detection or bot identification.
  • Replacing the cosine schedule with alternative noise schedules might reveal whether the reported gains depend on that specific choice.
  • If the contrastive term proves essential, similar stabilization could be added to existing graph anomaly detectors without redesigning their entire architecture.

Load-bearing premise

That noise-perturbed node-feature views created on a cosine schedule and stabilized by contrastive learning will reduce representation dilution for camouflaged fraud nodes without introducing new artifacts that hurt the target detection task.

What would settle it

Running the same 1-percent-label protocol on the public benchmarks and finding that ADC-GNN shows no improvement over the listed baselines, or that ablating the diffusion-contrastive component removes all reported gains.

Figures

Figures reproduced from arXiv: 2606.28134 by Chao Hu, Heyuan Shi, Liming Liu, Mingfei Lu, Xingle Li, Yiwei Ge.

Figure 1
Figure 1. Figure 1: Illustration of the few-shot and label-imbalance challenges in real-world transaction networks. Few-shot: only a small portion of nodes are labeled, result￾ing in insufficient supervision. Label im￾balance: fraudulent nodes are often buried within clusters of normal nodes. (i) Spatial GNNs improve ro￾bustness via sampling, pruning, or relation-aware aggregation, yet they still rely on message passing that … view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of ADC-GNN. aligned with the graph’s low-frequency spectrum, whereas fraudulent nodes manifest as high-frequency anomalies that deviate from local structural con￾sistency, motivating the use of spectral analysis to expose subtle structural irregularities that are difficult to capture through local aggregation alone. 4. Methodology Our proposed ADC-GNN ( [PITH_FULL_IMAGE:figures/full_f… view at source ↗
Figure 3
Figure 3. Figure 3: Coarse ablation visualization on four datasets at a 1% training ratio. “ours–a” removes the attention branch, “ours–d–c” removes the diffusion-contrastive branch, and “ours–a–c” removes both branches. This figure summarizes the large-component effect, while [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity of ADC-GNN to the cosine-schedule offset and diffusion timestep [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Each subfigure visualizes the 2D projection of node embeddings at different dif [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The left plot tracks the evolution of contrastive margin and linear separability [PITH_FULL_IMAGE:figures/full_fig_p029_6.png] view at source ↗
Figure 3
Figure 3. Figure 3: 31 [PITH_FULL_IMAGE:figures/full_fig_p032_3.png] view at source ↗
read the original abstract

Graph-based fraud detection is essential for safeguarding large-scale transaction systems, where undetected anomalies may lead to substantial financial losses and security risks. Real-world fraud graphs pose two coupled challenges: sparse and imbalanced supervision, where verified fraudulent labels are scarce and heavily skewed toward benign accounts, and representation dilution, where spatial message passing may oversmooth camouflaged anomalies while spectral filters may suppress fraud-relevant mid- and high-frequency irregularities. To address these challenges, we propose ADC-GNN, short for Attention-guided Diffusion-Contrastive Graph Neural Network, a unified framework that combines diffusion-guided feature augmentation, contrastive representation learning, and multi-hop spectral attention for few-shot graph fraud detection. The diffusion component is formulated as a feature-space denoising augmentation mechanism rather than a full topology-generative graph diffusion model: it constructs noise-perturbed node-feature views under a cosine schedule and uses contrastive learning to stabilize node representations across perturbations. The spectral attention module further adaptively emphasizes fraud-relevant hop-level and relation-level cues. We evaluate ADC-GNN primarily on three public benchmarks and additionally report a proprietary real-world telecom transaction dataset with approximately 60,000 records as a private case study. Under the 1% training setting, ADC-GNN achieves consistent improvements over original graph fraud baselines and four protocol-consistent recent graph anomaly/fraud baselines on the public benchmarks. Additional analyses on split stability, training ratios, oversampling alternatives, module-level ablations, diffusion schedules, and runtime and memory-consumption comparisons further characterize the effective operating regime of ADC-GNN.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript proposes ADC-GNN, an Attention-guided Diffusion-Contrastive Graph Neural Network for few-shot graph fraud detection. It combines diffusion-guided feature augmentation (constructing noise-perturbed node-feature views under a cosine schedule, stabilized via contrastive learning), contrastive representation learning, and multi-hop spectral attention to address sparse/imbalanced supervision and representation dilution in camouflaged fraud nodes. The central claim is that, under a 1% training setting, ADC-GNN delivers consistent improvements over original graph fraud baselines and four protocol-consistent recent graph anomaly/fraud baselines on three public benchmarks, with additional results on a private ~60k-record telecom dataset and supporting analyses (split stability, training ratios, ablations, diffusion schedules, runtime/memory).

Significance. If the reported gains are reproducible with full quantitative detail and the diffusion-contrastive module is shown not to introduce artifacts that degrade fraud-specific signals, the work could provide a practical feature-space augmentation route for GNNs under extreme label sparsity without requiring topology generation. The explicit framing of diffusion as denoising augmentation rather than generative modeling is a clear design strength.

major comments (2)
  1. [Abstract] Abstract: the claim that ADC-GNN 'achieves consistent improvements' under the 1% training setting supplies no numerical metrics, statistical tests, error bars, exact baseline reproduction details, or data-split leakage safeguards; without these the central empirical claim cannot be evaluated and the support for superiority remains unverified.
  2. [Method (diffusion-guided feature augmentation)] Diffusion component description: the formulation of noise-perturbed node-feature views under a cosine schedule stabilized by contrastive learning is presented as mitigating representation dilution for camouflaged fraud nodes, yet no analysis demonstrates that the chosen perturbation distribution preserves (rather than averages away) fraud-relevant mid/high-frequency irregularities or avoids injecting spurious correlations that the subsequent spectral attention cannot correct; this assumption is load-bearing for the 1% label gains.
minor comments (2)
  1. [Abstract] The abstract refers to 'four protocol-consistent recent graph anomaly/fraud baselines' without naming them or citing their sources, which hinders immediate assessment of protocol consistency.
  2. [Experiments] The private telecom dataset is described only by record count (~60,000) with no further statistics on graph size, label distribution, or feature dimensionality, limiting evaluation of the case-study results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below, indicating revisions where the manuscript will be strengthened.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that ADC-GNN 'achieves consistent improvements' under the 1% training setting supplies no numerical metrics, statistical tests, error bars, exact baseline reproduction details, or data-split leakage safeguards; without these the central empirical claim cannot be evaluated and the support for superiority remains unverified.

    Authors: We agree that the abstract would be more informative with representative quantitative details. The full experimental section already reports metrics with standard deviations across multiple splits, baseline reproduction protocols, and safeguards against leakage. We will revise the abstract to include key performance deltas and a brief note on the evaluation protocol (5 random splits, 1% labeled data). revision: yes

  2. Referee: [Method (diffusion-guided feature augmentation)] Diffusion component description: the formulation of noise-perturbed node-feature views under a cosine schedule stabilized by contrastive learning is presented as mitigating representation dilution for camouflaged fraud nodes, yet no analysis demonstrates that the chosen perturbation distribution preserves (rather than averages away) fraud-relevant mid/high-frequency irregularities or avoids injecting spurious correlations that the subsequent spectral attention cannot correct; this assumption is load-bearing for the 1% label gains.

    Authors: The diffusion module operates strictly in feature space as a denoising-style augmentation, with the cosine schedule and contrastive stabilization chosen to retain discriminative signals; ablations in the manuscript already quantify the contribution of this module to the reported gains. We acknowledge that an explicit frequency-domain preservation study would further substantiate the design choice. We will add such an analysis (pre/post-augmentation spectral energy on fraud vs. benign nodes) in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external benchmark comparisons

full rationale

The paper introduces ADC-GNN as a composite framework (diffusion-guided augmentation under cosine schedule + contrastive stabilization + multi-hop spectral attention) and supports its claims solely via reported performance gains on public benchmarks under 1% labels versus external baselines. No equations, self-citations, or derivation steps are present that reduce a claimed prediction or uniqueness result to the paper's own fitted inputs or prior self-work by construction. The method is described as a practical combination rather than a closed mathematical derivation, and the reader's assessment of score 1.0 aligns with the absence of any load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on several design assumptions about how diffusion augmentation and contrastive learning interact with graph spectral attention to solve the stated problems; these are not derived from first principles but introduced as part of the proposed method.

free parameters (2)
  • cosine diffusion schedule hyperparameters
    The noise perturbation schedule is specified as cosine but its exact parameters are design choices that typically require selection or tuning on data.
  • contrastive loss weighting coefficient
    Balance between the contrastive objective and supervised loss is a free parameter needed to stabilize training.
axioms (2)
  • domain assumption Feature-space diffusion under cosine schedule functions as effective denoising augmentation that preserves fraud-relevant signals
    Invoked when the abstract states the diffusion component is formulated as feature-space denoising rather than topology generation.
  • domain assumption Multi-hop spectral attention can adaptively emphasize fraud-relevant cues while avoiding suppression of mid- and high-frequency irregularities
    Central to the description of the spectral attention module.

pith-pipeline@v0.9.1-grok · 5820 in / 1482 out tokens · 37187 ms · 2026-06-29T04:55:52.751796+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

51 extracted references · 12 canonical work pages · 4 internal anchors

  1. [1]

    X. Ma, J. Wu, S. Xue, J. Yang, C. Zhou, Q. Z. Sheng, H. Xiong, L. Akoglu, A comprehensive survey on graph anomaly detection with deep learning, IEEE transactions on knowledge and data engineering 35 (12) (2021) 12012–12038

    2021

  2. [2]

    H. Qiao, Q. Wen, X. Li, E.-P. Lim, G. Pang, Generative semi-supervised graph anomaly detection, in: Advances in Neural Information Process- ing Systems, Vol. 37, 2024, pp. 4660–4688

    2024

  3. [3]

    Hamilton, Z

    W. Hamilton, Z. Ying, J. Leskovec, Inductive representation learning on large graphs, Advances in neural information processing systems 30 (2017)

    2017

  4. [4]

    K. Xu, W. Hu, J. Leskovec, S. Jegelka, How powerful are graph neural networks?, arXiv preprint arXiv:1810.00826 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  5. [5]

    Graph Attention Networks

    P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Lio, Y. Bengio, Graph attention networks, arXiv preprint arXiv:1710.10903 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  6. [6]

    F. Xu, N. Wang, H. Wu, X. Wen, X. Zhao, H. Wan, Revisiting graph- based fraud detection in sight of heterophily and spectrum, in: Proceed- ings of the AAAI conference on artificial intelligence, Vol. 38, 2024, pp. 9214–9222. 36

    2024

  7. [7]

    B. Fang, H. Chen, W. Wang, Y. Wang, Graphfa: Graph enhanced fraud detectors with camouflage detection for financial anti-fraud, in: 2024 9th International Conference on Intelligent Computing and Signal Pro- cessing (ICSP), IEEE, 2024, pp. 323–327

    2024

  8. [8]

    T. N. Kipf, M. Welling, Semi-supervised classification with graph con- volutional networks, arXiv preprint arXiv:1609.02907 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  9. [9]

    Adap- tive universal generalized pagerank graph neural net- work,

    E. Chien, J. Peng, P. Li, O. Milenkovic, Adaptive universal generalized pagerankgraphneuralnetwork, arXivpreprintarXiv:2006.07988(2020)

    work page arXiv 2006

  10. [10]

    J. Tang, F. Hua, Z. Gao, P. Zhao, J. Li, Gadbench: Revisiting and benchmarking supervised graph anomaly detection, Advances in Neural Information Processing Systems 36 (2023) 29628–29653

    2023

  11. [11]

    X. Li, L. Chen, Graph anomaly detection with domain-agnostic pre- training and few-shot adaptation, in: 2024 IEEE 40th International Conference on Data Engineering (ICDE), IEEE, 2024, pp. 2667–2680

    2024

  12. [12]

    Rayana, L

    S. Rayana, L. Akoglu, Collective opinion spam detection: Bridging re- view networks and metadata, in: Proceedings of the 21th acm sigkdd international conference on knowledge discovery and data mining, 2015, pp. 985–994

    2015

  13. [13]

    J. J. McAuley, J. Leskovec, From amateurs to connoisseurs: modeling the evolution of user expertise through online reviews, in: Proceedings of the 22nd international conference on World Wide Web, 2013, pp. 897–908

    2013

  14. [14]

    J. Tang, J. Li, Z. Gao, J. Li, Rethinking graph neural networks for anomaly detection, in: International conference on machine learning, PMLR, 2022, pp. 21076–21089

    2022

  15. [15]

    Y. Dou, Z. Liu, L. Sun, Y. Deng, H. Peng, P. S. Yu, Enhancing graph neural network-based fraud detectors against camouflaged fraudsters, in: Proceedings of the 29th ACM international conference on information & knowledge management, 2020, pp. 315–324

    2020

  16. [16]

    Zhang, J

    G. Zhang, J. Wu, J. Yang, A. Beheshti, S. Xue, C. Zhou, Q. Z. Sheng, Fraudre: Fraud detection dual-resistant to graph inconsistency 37 and imbalance, in: 2021 IEEE international conference on data mining (ICDM), IEEE, 2021, pp. 867–876

    2021

  17. [17]

    Y. Liu, X. Ao, Z. Qin, J. Chi, J. Feng, H. Yang, Q. He, Pick and choose: a gnn-based imbalanced learning approach for fraud detection, in: Proceedings of the web conference 2021, 2021, pp. 3168–3177

    2021

  18. [18]

    Xiang, M

    S. Xiang, M. Zhu, D. Cheng, E. Li, R. Zhao, Y. Ouyang, L. Chen, Y. Zheng, Semi-supervised credit card fraud detection via attribute- driven graph representation, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 37, 2023, pp. 14557–14565

    2023

  19. [19]

    Zhang, Z

    J. Zhang, Z. Xu, D. Lv, Z. Shi, D. Shen, J. Jin, F. Dong, Dig-in-gnn: discriminative feature guided gnn-based fraud detector against incon- sistencies in multi-relation fraud graph, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38, 2024, pp. 9323–9331

    2024

  20. [20]

    N. Chen, Z.Liu, B.Hooi, B.He, R.Fathony, J.Hu, J.Chen, Consistency training with learnable data augmentation for graph anomaly detection with limited supervision, In The Twelfth International Conference on Learning Representations (ICLR 2024), spotlight paper (2024). URLhttps://openreview.net/forum?id=elMKXvhhQ9

    2024

  21. [21]

    Cheng, X

    D. Cheng, X. Wang, Y. Zhang, L. Zhang, Graph neural network for fraud detection via spatial-temporal attention, IEEE Transactions on Knowledge and Data Engineering 34 (8) (2020) 3800–3813

    2020

  22. [22]

    S. Tian, J. Dong, J. Li, W. Zhao, X. Xu, B. Song, C. Meng, T. Zhang, L. Chen, et al., Sad: Semi-supervised anomaly detection on dynamic graphs, arXiv preprint arXiv:2305.13573 (2023)

    work page arXiv 2023

  23. [23]

    B. Xu, H. Shen, Q. Cao, Y. Qiu, X. Cheng, Graph wavelet neural net- work, arXiv preprint arXiv:1904.07785 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  24. [24]

    Z. Chai, S. You, Y. Yang, S. Pu, J. Xu, H. Cai, W. Jiang, Can ab- normality be detected by graph neural networks?, in: IJCAI, 2022, pp. 1945–1951

    2022

  25. [25]

    B. Wu, X. Yao, B. Zhang, K.-M. Chao, Y. Li, Splitgnn: Spectral graph neural network for fraud detection against heterophily, in: Proceedings 38 ofthe32ndACMinternationalconferenceoninformationandknowledge management, 2023, pp. 2737–2746

    2023

  26. [26]

    R. Guo, M. Zou, S. Zhang, X. Zhang, Z. Yu, Z. Feng, Graph local ho- mophily network for anomaly detection, in: Proceedings of the 33rd ACM International Conference on Information and Knowledge Manage- ment, 2024, pp. 706–716

    2024

  27. [27]

    J. Tang, H. Gu, D. B. Vuković, G. Xu, Y. Wang, H. Tao, J. Cao, Fraud detection in multi-relation graph: Contrastive learning on feature and structural levels, Neurocomputing 637 (2025) 130063

    2025

  28. [28]

    J. Liu, Y. Tian, G. Liu, Global confidence degree based graph neural network for financial fraud detection, arXiv preprint arXiv:2407.17333 (2024)

    work page arXiv 2024

  29. [29]

    Y. Gao, X. Wang, X. He, Z. Liu, H. Feng, Y. Zhang, Addressing het- erophily in graph anomaly detection: A perspective of graph spectrum, in: Proceedings of the ACM web conference 2023, 2023, pp. 1528–1538

    2023

  30. [30]

    Zhang, X

    Y. Zhang, X. Ma, J. Wu, J. Yang, H. Fan, Heterogeneous subgraph transformer for fake news detection, in: Proceedings of the ACM Web Conference 2024, 2024, pp. 1272–1282

    2024

  31. [31]

    Z. Liu, C. Chen, X. Yang, J. Zhou, X. Li, L. Song, Heterogeneous graph neural networks for malicious account detection, in: Proceedings of the 27th ACM international conference on information and knowledge man- agement, 2018, pp. 2077–2085

    2018

  32. [32]

    S. Li, B. Qiao, K. Li, Q. Lu, M. Lin, W. Zhou, Multi-modal social bot detection: Learning homophilic and heterophilic connections adaptively, in: Proceedings of the 31st ACM International Conference on Multime- dia, 2023, pp. 3908–3916

    2023

  33. [33]

    M. Duan, T. Zheng, Y. Gao, G. Wang, Z. Feng, X. Wang, Dga-gnn: Dynamic grouping aggregation gnn for fraud detection, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38, 2024, pp. 11820–11828

    2024

  34. [34]

    G. Hu, Y. Liu, Q. He, X. Ao, F2gnn: An adaptive filter with feature segmentation for graph-based fraud detection, in: ICASSP 2024-2024 39 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP), IEEE, 2024, pp. 6335–6339

    2024

  35. [35]

    Y.Wan, D.Zhang, D.Liu, F.Xiao, CGAD:Anovelcontrastivelearning- based framework for anomaly detection in attributed networks, Neuro- computing 609 (2024) 128379.doi:10.1016/j.neucom.2024.128379

    work page doi:10.1016/j.neucom.2024.128379 2024

  36. [36]

    Y. Liu, S. Li, Y. Zheng, Q. Chen, C. Zhang, S. Pan, ARC: A generalist graph anomaly detector with in-context learning, in: Advances in Neural Information Processing Systems, Vol. 37, 2024, pp. 50772–50804

    2024

  37. [37]

    Y. Lin, J. Tang, C. Zi, H. V. Zhao, Y. Yao, J. Li, UniGAD: Unifying multi-levelgraphanomalydetection, in: AdvancesinNeuralInformation Processing Systems, Vol. 37, 2024, pp. 136120–136148

    2024

  38. [38]

    P. Li, H. Yu, X. Luo, Context-aware graph neural network for graph- based fraud detection with extremely limited labels, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39, 2025, pp. 12112–12120.doi:10.1609/aaai.v39i11.33319

    work page doi:10.1609/aaai.v39i11.33319 2025

  39. [39]

    C. Li, M. Lin, Z. Ding, N. Lin, Y. Zhuang, Y. Huang, X. Ding, L. Cao, Knowledge condensation distillation, in: European Conference on Com- puter Vision, Springer, 2022, pp. 19–35

    2022

  40. [40]

    X. Shen, Q. Dai, S. Mao, F.-l. Chung, K.-S. Choi, Network together: Node classification via cross-network deep network embedding, arXiv preprint arXiv:1901.07264 (2019)

    work page arXiv 1901

  41. [41]

    H. Xu, Y. Zhang, L. Sun, C. Li, Y. Huang, X. Ding, AFSC: Adap- tive fourier space compression for anomaly detection, arXiv preprint arXiv:2204.07963 (2022)

    work page arXiv 2022

  42. [42]

    Kholkin, I

    S. Kholkin, I. Butakov, E. Burnaev, N. Gushchin, A. Korotin, In- foBridge: Mutual information estimation via bridge matching, arXiv preprint arXiv:2502.01383 (2025)

    work page arXiv 2025

  43. [43]

    C. Li, X. Liu, C. Wang, Y. Liu, W. Yu, J. Shao, Y. Yuan, GTP- 4o: Modality-prompted heterogeneous graph learning for omni-modal biomedical representation, in: Computer Vision – ECCV 2024, Springer, 2024, pp. 168–187. 40

    2024

  44. [44]

    A. Q. Nichol, P. Dhariwal, Improved denoising diffusion probabilistic models, in: International conference on machine learning, PMLR, 2021, pp. 8162–8171

    2021

  45. [45]

    Dhariwal, A

    P. Dhariwal, A. Nichol, Diffusion models beat gans on image synthesis, Advances in neural information processing systems 34 (2021) 8780–8794

    2021

  46. [46]

    Glorot, A

    X. Glorot, A. Bordes, Y. Bengio, Deep sparse rectifier neural networks, in: Proceedings of the fourteenth international conference on artificial intelligence and statistics, JMLR Workshop and Conference Proceed- ings, 2011, pp. 315–323

    2011

  47. [47]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov, Dropout: A simple way to prevent neural networks from overfitting, Journal of Machine Learning Research 15 (56) (2014) 1929–1958

    2014

  48. [48]

    Defferrard, X

    M. Defferrard, X. Bresson, P. Vandergheynst, Convolutional neural net- works on graphs with fast localized spectral filtering, Advances in neural information processing systems 29 (2016)

    2016

  49. [49]

    J. Hu, L. Shen, G. Sun, Squeeze-and-excitation networks, in: Proceed- ings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  50. [50]

    Bhatia, B

    S. Bhatia, B. Hooi, M. Yoon, K. Shin, C. Faloutsos, Midas: Microcluster-based detector of anomalies in edge streams, in: Proceed- ings of the AAAI Conference on Artificial Intelligence, Vol. 34, 2020, pp. 3242–3249

    2020

  51. [51]

    T. Chen, S. Kornblith, M. Norouzi, G. Hinton, A simple framework for contrastive learning of visual representations, in: International confer- ence on machine learning, PmLR, 2020, pp. 1597–1607. 41

    2020