Pith. sign in

REVIEW 5 major objections 7 minor 66 references

REGE: A Method for Incorporating Uncertainty in Graph Embeddings

T0 review · 5 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read REGE incorporates per-node uncertainty radii into graph embedding training and reports average accuracy gains of 1.5% over eight state-of-the-art defenses under structural attacks on four datasets.

desk verdict A plausible new uncertainty-based defense whose headline accuracy gain is overstated and whose attack evaluation needs adaptive baselines; worth a serious revision. read the letter →

arxiv 2412.05735 v1 pith:2YDJFWB2 submitted 2024-12-07 cs.LG

Zohair Shafi , Germans Savcisens , Tina Eliassi-Rad This is my paper · ORCID
classification cs.LG
keywords graphembeddingsuncertaintyquantificationadversarialrobustnessneuralnetworksconformalpredictioncurriculumlearningeigen-decompositionnodeclassification
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 argues that graph embeddings should carry an explicit per-node uncertainty radius, and that training with noise proportional to that radius makes node classification more robust to adversarial attacks on the graph structure. It proposes REGE, which derives two kinds of radii: data-dependent radii from low-rank eigen-reconstructions of the adjacency matrix, and model-dependent radii from a student-teacher setup with conformal quantile regression. It then trains a GCN with curriculum learning on progressively fuller graph reconstructions while injecting Gaussian noise with variance equal to the radius into hidden layers. Across Cora, Citeseer, CoraML, and PolBlogs, under MinMax, Meta-Attack, and GraD perturbations of 1–10%, REGE outperforms eight existing defenses by an average of 1.5% accuracy.

What carries the argument

The central object is the per-node uncertainty radius r_i, paired with the rule that hidden representations are perturbed as x_i^l ← x_i^l + N(0, r_i). The radius is produced by two procedures: the data-dependent radius (DDR) averages, row-wise, the binary deviation of entries in a consensus weighted adjacency matrix built from low-rank eigen-reconstructions of the graph; the model-dependent radius (MDR) conformalizes student-teacher quantile intervals across embedding dimensions. These radii are then used both as noise variances and as the curriculum schedule, since the graph views with few components are treated as easy examples and more components are added progressively.

What would settle it

An experiment that replaces each node's radius with a random value in [0,1] (or with the same constant) while keeping the curriculum schedule unchanged; if the accuracy under MinMax, Meta-Attack, and GraD remains within 0.5% of REGE with true radii, then the specific radius values are not what drives the reported improvement.

Watch

Extended reading notes

Core claim

REGE's central claim is that measuring and incorporating uncertainty into graph embedding training yields embeddings that are more resistant to adversarial structural perturbations. The method computes, for each node i, a scalar radius r_i in [0,1] from either the data (DDR) or the model (MDR). DDR uses the eigen-decomposition of the adjacency matrix to generate graph views with increasing numbers of components, averages them into a weighted adjacency matrix W, and maps each row through a binary deviation function to a node-level radius. MDR trains a teacher GCN, then a student MLP with quantile loss to predict per-dimension quantiles of the teacher's embeddings, conformalizes those intervals using the training set as calibration, and averages the dimensional interval widths into a radius. During training, Gaussian noise N(0, r_i) is added to each node's hidden representation, and the model is trained with curriculum learning on the sequence of graph views. The paper reports that this consistently improves node classification accuracy under MinMax, Meta-Attack, and GraD perturbations, by an average of 1.5% against state-of-the-art defenses.

Load-bearing premise

The load-bearing premise is that the computed radius values genuinely measure node uncertainty, so that adding Gaussian noise to hidden representations with variance equal to the radius is what improves robustness, rather than the noise or the curriculum alone.

Editorial extensions

If this is right

  • If REGE's central claim holds, any downstream task that uses the produced embeddings inherits a measurable per-node uncertainty that can flag unreliable predictions.
  • Adding radius-proportional noise during training is a defense that requires no separate attack model, unlike adversarial training.
  • Because the radii are computed from an eigen-decomposition and a teacher GCN, the same recipe can be attached to any differentiable graph embedding function, not just GCNs.
  • The 1.5% average accuracy gain is consistent across perturbation levels from 1% to 10% and across three attack families, suggesting the defense is not attack-specific.
  • Curriculum learning over graph views with increasing eigen-components contributes beyond the noise injection: the REGE-NCT ablation (no curriculum) underperforms full REGE in most cases.

Reading between the lines

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

  • A natural next test is whether the radius values carry meaning beyond the training objective: for instance, whether nodes with large MDR correspond to misclassified or low-degree nodes in other graph tasks, which would make REGE a source of calibrated confidence scores as well as a defense.
  • The conformal step in REGE is applied with the training set as calibration, which breaks the standard distribution-free coverage guarantee; a held-out calibration set would be needed to claim that the model-dependent radii are statistically calibrated.
  • If the gains come mainly from the curriculum over low-rank views rather than from the radius noise, then a simpler method that trains on a few low-rank views without radii might match REGE; the paper isolates curriculum but not the radii, so this decomposition is an open empirical question.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. The paper introduces REGE, a training-time method for graph neural networks that computes per-node uncertainty radii from two sources: (i) data-dependent radii (DDR) obtained by averaging a consensus of low-rank eigen-reconstructions of the adjacency matrix and applying a 'binary deviation' function; and (ii) model-dependent radii (MDR) obtained from a student-teacher setup in which a student MLP predicts quantiles of a teacher GCN's output embeddings, refined by conformal quantile regression. These radii are used to inject Gaussian noise into hidden representations (Eq. 2.9), and training proceeds with curriculum learning on the sequence of graph views. The paper evaluates REGE on node classification under MinMax, Meta-Attack, and GraD structural attacks on Cora, Citeseer, and PolBlogs (plus CoraML in the appendix), reporting that REGE outperforms eight defense baselines by an average of 1.5% accuracy.

Significance. If the robustness gains are real and the radii genuinely reflect node-level uncertainty, REGE would be a simple and inexpensive addition to GCN training. The method is straightforward to implement, the code is released, and the idea of injecting uncertainty-scaled noise into hidden layers is plausible. However, the current evaluation does not establish the central claim: the attack protocol appears to use surrogate-generated perturbations rather than adaptive attacks, the conformal calibration is performed on the training set, which voids the stated coverage guarantee, and the ablation study does not isolate the effect of the radius noise. These gaps currently prevent the reader from attributing the reported improvements to the proposed mechanism.

major comments (5)
  1. [Section 3.1, Tables 1-3] The paper does not state whether the MinMax, Meta-Attack, and GraD perturbations are optimized against each defense or generated once with a surrogate GCN and then reused for all methods. The sentence 'adversarial graphs for Meta-Attack were computed using a standard GCN surrogate model' suggests the latter for at least one attack. If all defenses are tested on the same surrogate-attacked graphs, the evaluation measures performance on one fixed perturbation set, not robustness to attacks tailored to each defense, which is a known failure mode for non-adaptive evaluations (Athalye et al., 2018). The authors should specify the protocol for all three attacks and add at least one adaptive attack (e.g., attacking the final REGE model with the same budget) to support the central robustness claim.
  2. [Section 2.2, Eq. (2.5)-(2.7)] The text first says conformal calibration requires a set 'different from validation or testing data,' then states that conformal learning is applied 'using the training set as the calibration set.' Since the student MLP is trained on the teacher's outputs, using the training set for calibration violates the exchangeability requirement of conformal prediction, so the MDR intervals have no stated coverage guarantee. The authors should use a held-out calibration set or revise the claim that the intervals provide guaranteed coverage.
  3. [Section 3.3, Table 4] The ablation study only removes curriculum learning (REGE-NCT); every variant still injects radius-scaled noise. To support the claim that incorporating uncertainty improves robustness, the paper needs a control without the noise (e.g., GCN with curriculum learning only, or with uniform/random noise) to show that the specific radius values matter. Without this control, the improvement might be attributable to curriculum learning or to noise injection in general.
  4. [Tables 1-6] The paper reports 10-run means with standard deviations but no paired significance tests. Many of the reported gains are within one standard deviation of the best baseline (e.g., Table 2, Cora MinMax 1%: REGE-D .837±.004 vs GraphReshape .834±.003; Table 1, Citeseer GraD 10%: REGE-M .696±.057 vs GADC .701±.004 is a loss; Table 3, PolBlogs GraD 10%: REGE-D .865 ties GCN-SVD and Ricci-GNN; Table 6, Citeseer GraD 3%: REGE-M .694 vs GADC .696 loses). The headline 'average 1.5%' therefore lacks statistical support and should be accompanied by paired tests or bootstrap intervals, along with a per-condition breakdown of gains and losses.
  5. [Abstract and Conclusion] The abstract claims an average improvement of 1.5% without specifying the exact settings, and the conclusion says 'across four datasets,' but the main experimental section evaluates only three datasets (Cora, Citeseer, PolBlogs); CoraML appears only in the appendix. The paper should specify exactly which tables and settings are averaged, report the per-condition gains/losses, and correct the dataset count.
minor comments (7)
  1. [Section 2.1, Eq. (2.2)] The binary deviation function is one of many possible edge-uncertainty measures; Figure 2 gives a qualitative comparison but no quantitative ablation of alternative functions (e.g., standard deviation or entropy). Consider adding such an ablation to justify the choice.
  2. [Section 2.2] The student MLP predicts the teacher GCN's embedding outputs, but the paper does not specify how the teacher is trained (e.g., the loss and the label splits) or how the student's quantile predictions are converted into a final radius after averaging over dimensions. A precise description would improve reproducibility.
  3. [Section 2.3, Eq. (2.9)] The noise variance is set to the radius ri, but the paper does not discuss the range of MDR values or whether clipping is needed; if radii can exceed 1, the noise may overwhelm the hidden representations.
  4. [Algorithm 2 and Section A.5] Algorithm 2 iterates k over q, q+1, ..., n, while the appendix says components are incremented by 5; these should be made consistent.
  5. [Section 3.1] The dataset splits and the number of labeled examples per class are not reported, which makes the node classification setup difficult to reproduce.
  6. [Typos] Section 2.1 contains a duplicated phrase ('uncertainty arising fr In this paper'); Table 5 caption has 'consistenly' and Table 3 caption has 'consistenly' as well.
  7. [Related Work] UAG (Feng et al., 2021) is discussed in Related Work but is not included in the experimental comparisons; given that UAG also uses uncertainty for robustness, it would be a relevant baseline.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: REGE's radii are empirical inputs, and its robustness claim is an experimentally measured accuracy gain, not a quantity fitted or defined into existence.

full rationale

The claimed result—that REGE improves node-classification accuracy under adversarial attacks by an average of 1.5%—is an empirical comparison against external defense baselines on attacked graphs. Neither the data-dependent radii (Eqs. 2.3–2.4, computed from eigen-reconstructions of the clean adjacency matrix) nor the model-dependent radii (student-teacher quantile intervals refined by conformal learning) are fitted to the adversarial-accuracy target; they are fixed inputs used to inject training noise (Eq. 2.9) and to order curriculum views. No equation in the paper reduces the reported accuracy to the radius values by construction, and no fitted parameter is renamed as a prediction. The method relies on external works for spectral reconstruction, distillation, conformal prediction, and curriculum learning; the only same-author citation [2] appears in a domain-example list and is not load-bearing. The use of the training set as the conformal calibration set is a validity flaw in the stated coverage guarantee, and the reliance on surrogate-generated attacks and absence of significance tests are experimental-design concerns, but neither is a circular reduction. The derivation chain is therefore self-contained with respect to circularity.

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

REGE does not introduce new physical or mathematical entities; it defines new scalar quantities (data- and model-dependent radii) that are functions of the existing graph and model. The central claim rests on several untested assumptions about what the radii mean and why noise injection should help, plus hand-chosen parameters q, increment, threshold, and alpha.

free parameters (5)
  • minimum eigen-components q
    Input to Algorithm 1 that determines the first and simplest curriculum view. No default value is given in the main text or appendix.
  • component increment step = 5
    Appendix A.5: 'we increment a number of components by 5 as we observe no performance gain from finer increments.' This is a hand-chosen value without reported sensitivity analysis.
  • discretization threshold = 0.5
    Appendix A.5: reconstructed adjacency matrices are scaled to [0,1] with a threshold of 0.5 to discretize. This threshold determines which edges appear in each view and directly affects the radii.
  • conformal coverage alpha = 0.05
    Section 2.2 states 'We set α = 0.05.' This controls the quantile intervals used to compute model-dependent radii.
  • training epochs per view = 100
    Appendix A.5: 'we train on each graph Gi for 100 epochs.' A standard but hand-chosen training budget.
assumptions (5)
  • domain assumption Low-rank eigen-reconstructions of a graph preserve high-certainty edges better than low-certainty edges, so a consensus over views yields edge uncertainty.
    Section 2.1: 'reconstructing a graph from eigenvectors associated with large eigenvalues reconstructs a graph that has edges with high certainty', citing references [15,16,17].
  • ad hoc to paper The binary deviation function u_e = 1 - |W_ij - (1 - W_ij)| is an appropriate measure of edge uncertainty.
    Equation 2.2 in Section 2.1. The only justification is that W=0.5 gives maximum uncertainty; no theoretical or empirical validation is provided.
  • ad hoc to paper Injecting Gaussian noise with variance equal to a node's radius into hidden layer representations improves robustness.
    Section 2.3, Equation 2.9. The paper likens this to VAEs but does not prove or isolate why radius-scaled noise helps; the ablation only removes curriculum, not the noise scheme.
  • domain assumption Training on progressively higher-rank graph reconstructions is an effective curriculum for robustness.
    Section 2.3 and Algorithm 2, citing curriculum learning [8]. The benefit is shown in Table 4 but no mechanism is established.
  • ad hoc to paper Conformal quantile regression with calibration on the training set provides valid prediction intervals.
    Section 2.2: 'Conformal learning is then used to refine these intervals (using the training set as the calibration set).' This violates the exchangeability requirement for conformal coverage; the stated statistical guarantee does not hold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of REGE: A Method for Incorporating Uncertainty in Graph Embeddings." pith.science (2026). https://pith.science/paper/2YDJFWB2

@misc{pith2026241205735,
  author       = {Pith},
  title        = {Pith review of: REGE: A Method for Incorporating Uncertainty in Graph Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2YDJFWB2}},
  note         = {Machine review of arXiv:2412.05735}
}
read the original abstract

Machine learning models for graphs in real-world applications are prone to two primary types of uncertainty: (1) those that arise from incomplete and noisy data and (2) those that arise from uncertainty of the model in its output. These sources of uncertainty are not mutually exclusive. Additionally, models are susceptible to targeted adversarial attacks, which exacerbate both of these uncertainties. In this work, we introduce Radius Enhanced Graph Embeddings (REGE), an approach that measures and incorporates uncertainty in data to produce graph embeddings with radius values that represent the uncertainty of the model's output. REGE employs curriculum learning to incorporate data uncertainty and conformal learning to address the uncertainty in the model's output. In our experiments, we show that REGE's graph embeddings perform better under adversarial attacks by an average of 1.5% (accuracy) against state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2412.05735 by the authors.

Figure 1
Figure 1. REGE uses eigen-decomposition to generate multiple graph views by systematically reconstructing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Radii for each node in the PolBlogs network, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Low degree nodes (bold) have low DDR possi￾bly due to consistent edge reconstruction but high MDR indicating graph embedding functions may struggle to learn robust representations for low degree nodes. graph autoencoder (VGAE) and a discriminator net￾work to learn robust representations. Zhang et al. [63] also use VGAE to reconstruct an attacked graph, fol￾lowed by using a vanilla GCN on the reconstructed graph. He … view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) Visualization of the Karate Club network with (b) data-dependent and (c) model-dependent radii. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Comparison between REGE and GCN-SVD under varying numbers of components for datasets [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 54 canonical work pages

  1. [1]

    A review on graph neural network methods in financial applica- tions,

    J. Wang, S. Zhang, Y. Xiao, and R. Song, “A review on graph neural network methods in financial applica- tions,” arXiv preprint arXiv:2111.15367 , 2021

    arXiv 2021

  2. [2]

    Improving the general- izability of protein-ligand binding predictions with ai- bind,

    A. Chatterjee, R. Walters, Z. Shafi, O. S. Ahmed, M. Sebek, D. Gysi, R. Yu, T. Eliassi-Rad, A.-L. Barab´ asi, and G. Menichetti, “Improving the general- izability of protein-ligand binding predictions with ai- bind,” Nature Communications, vol. 14, no. 1, p. 1989, 2023

    work page 1989

  3. [3]

    Hyganno: hybrid graph neural network– based cell type annotation for single-cell atac sequenc- ing data,

    W. Zhang, Y. Cui, B. Liu, M. Loza, S.-J. Park, and K. Nakai, “Hyganno: hybrid graph neural network– based cell type annotation for single-cell atac sequenc- ing data,” Briefings in Bioinformatics , vol. 25, no. 3, p. bbae152, 2024

    work page 2024

  4. [4]

    Graph convolutional neu- ral networks for web-scale recommender systems,

    R. Ying, R. He, K. Chen, P. Eksombatchai, W. L. Hamilton, and J. Leskovec, “Graph convolutional neu- ral networks for web-scale recommender systems,” in ACM SIGKDD, 2018, pp. 974–983

    work page 2018

  5. [5]

    Distilling the knowledge in a neural net- work,

    G. Hinton, “Distilling the knowledge in a neural net- work,” arXiv preprint arXiv:1503.02531 , 2015

    arXiv 2015

  6. [6]

    Graph-less neural networks: Teaching old mlps new tricks via distillation,

    S. Zhang, Y. Liu, Y. Sun, and N. Shah, “Graph-less neural networks: Teaching old mlps new tricks via distillation,” in ICLR, 2022

    work page 2022

  7. [7]

    A gentle intro- duction to conformal prediction and distribution- free uncertainty quantification,

    A. N. Angelopoulos and S. Bates, “A gentle intro- duction to conformal prediction and distribution- free uncertainty quantification,” arXiv preprint arXiv:2107.07511, 2021

    arXiv 2021

  8. [8]

    Curriculum learning,

    Y. Bengio, J. Louradour, R. Collobert, and J. Weston, “Curriculum learning,” in ICML, 2009, pp. 41–48

    work page 2009

Show all 66 references
  1. [9]

    Topology attack and defense for graph neural networks: an optimization perspective,

    K. Xu, H. Chen, S. Liu, P.-Y. Chen, T.-W. Weng, M. Hong, and X. Lin, “Topology attack and defense for graph neural networks: an optimization perspective,” in IJCAI, 2019, pp. 3961–3967

    2019

  2. [10]

    Adversarial attacks on graph neural networks via meta learning,

    D. Z¨ ugner and S. G¨ unnemann, “Adversarial attacks on graph neural networks via meta learning,” in ICLR, 2019

    2019

  3. [11]

    Towards reasonable budget allocation in untargeted graph struc- ture attacks via gradient debias,

    Z. Liu, Y. Luo, L. Wu, Z. Liu, and S. Z. Li, “Towards reasonable budget allocation in untargeted graph struc- ture attacks via gradient debias,” NeurIPS, pp. 27 966– 27 977, 2022

    2022

  4. [12]

    Bayesian inference of network structure from unreli- able data,

    J.-G. Young, G. T. Cantwell, and M. Newman, “Bayesian inference of network structure from unreli- able data,” Journal of Complex Networks , vol. 8, 2020

    2020

  5. [13]

    Network reconstruction via the min- imum description length principle,

    T. P. Peixoto, “Network reconstruction via the min- imum description length principle,” arXiv preprint arXiv:2405.01015, 2024

    2024 arXiv

  6. [14]

    The minimum description length principle in coding and modeling,

    A. Barron, J. Rissanen, and B. Yu, “The minimum description length principle in coding and modeling,” IEEE Transactions on Information Theory , vol. 44, 1998

    1998

  7. [15]

    Fast attributed graph embedding via density of states,

    S. Sawlani, L. Zhao, and L. Akoglu, “Fast attributed graph embedding via density of states,” in ICDM. IEEE, 2021, pp. 559–568

    2021

  8. [16]

    Localization on low-order eigenvectors of data matrices,

    M. Cucuringu and M. W. Mahoney, “Localization on low-order eigenvectors of data matrices,” arXiv preprint arXiv:1109.1355, 2011

    2011 arXiv

  9. [17]

    Spectral and dy- namical properties in classes of sparse networks with mesoscopic inhomogeneities,

    M. Mitrovi´ c and B. Tadi´ c, “Spectral and dy- namical properties in classes of sparse networks with mesoscopic inhomogeneities,” Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , vol. 80, no. 2, p. 026123, 2009

    2009

  10. [18]

    The political blogo- sphere and the 2004 us election: divided they blog,

    L. A. Adamic and N. Glance, “The political blogo- sphere and the 2004 us election: divided they blog,” in ACM LinkKDD, 2005, pp. 36–43

    2004

  11. [19]

    A survey on epistemic (model) uncertainty in su- pervised learning: Recent advances and applications,

    X. Zhou, H. Liu, F. Pourpanah, T. Zeng, and X. Wang, “A survey on epistemic (model) uncertainty in su- pervised learning: Recent advances and applications,” Neurocomputing, vol. 489, pp. 449–465, 2022

    2022

  12. [20]

    Aleatoric and epis- temic uncertainty in machine learning: An introduc- tion to concepts and methods,

    E. H¨ ullermeier and W. Waegeman, “Aleatoric and epis- temic uncertainty in machine learning: An introduc- tion to concepts and methods,” Machine Learning, vol. 110, no. 3, pp. 457–506, 2021

    2021

  13. [21]

    Auto-encoding variational bayes,

    D. P. Kingma, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114 , 2013

    2013 arXiv

  14. [22]

    Automating the construction of internet portals with machine learning,

    A. K. McCallum, K. Nigam, J. Rennie, and K. Sey- more, “Automating the construction of internet portals with machine learning,” Information Retrieval, vol. 3, pp. 127–163, 2000

    2000

  15. [23]

    Cite- seer: An automatic citation indexing system,

    C. L. Giles, K. D. Bollacker, and S. Lawrence, “Cite- seer: An automatic citation indexing system,” in ACM Conference on Digital Libraries , 1998, pp. 89–98

    1998

  16. [24]

    Robust graph convolutional networks against adversarial attacks,

    D. Zhu, Z. Zhang, P. Cui, and W. Zhu, “Robust graph convolutional networks against adversarial attacks,” in ACM SIGKDD, 2019, pp. 1399–1407

    2019

  17. [25]

    All you need is low (rank) defend- ing against adversarial attacks on graphs,

    N. Entezari, S. A. Al-Sayouri, A. Darvishzadeh, and E. E. Papalexakis, “All you need is low (rank) defend- ing against adversarial attacks on graphs,” in ACM WSDM, 2020, pp. 169–177

    2020

  18. [26]

    Graph structure learning for robust graph neural networks,

    W. Jin, Y. Ma, X. Liu, X. Tang, S. Wang, and J. Tang, “Graph structure learning for robust graph neural networks,” in ACM SIGKDD, 2020, pp. 66–74

    2020

  19. [27]

    Gnnguard: Defending graph neural networks against adversarial attacks,

    X. Zhang and M. Zitnik, “Gnnguard: Defending graph neural networks against adversarial attacks,” NeurIPS, pp. 9263–9275, 2020

    2020

  20. [28]

    Graph structure reshaping against adversarial attacks on graph neural networks,

    H. Wang, C. Zhou, X. Chen, J. Wu, S. Pan, Z. Li, J. Wang, and S. Y. Philip, “Graph structure reshaping against adversarial attacks on graph neural networks,” IEEE TKDE , 2024

    2024

  21. [29]

    Graph adversarial diffusion convolution,

    S. Liu, J. Chen, T. Fu, L. Lin, M. Zitnik, and D. Wu, “Graph adversarial diffusion convolution,” in ICML, 2024

    2024

  22. [30]

    Ricci-gnn: Defending against structural attacks through a geometric approach,

    Z. Ye, T. Ma, C.-C. Ni, K. S. Liu, J. Gao, and C. Chen, “Ricci-gnn: Defending against structural attacks through a geometric approach,” 2020

    2020

  23. [31]

    Diffusion improves graph learning,

    J. Gasteiger, S. Weißenberger, and S. G¨ unnemann, “Diffusion improves graph learning,” NeurIPS, vol. 32, 2019

    2019

  24. [32]

    Ricci curvature of markov chains on metric spaces,

    Y. Ollivier, “Ricci curvature of markov chains on metric spaces,” Journal of Functional Analysis , vol. 256, pp. 810–864, 2009

    2009

  25. [33]

    An information flow model for con- flict and fission in small groups,

    W. W. Zachary, “An information flow model for con- flict and fission in small groups,” Journal of Anthropo- logical Research, vol. 33, no. 4, pp. 452—-473, 1977

    1977

  26. [34]

    Dgcu: A new deep directed method based on gaussian embed- ding for clustering uncertain graphs,

    M. Danesh, M. Dorrigiv, and F. Yaghmaee, “Dgcu: A new deep directed method based on gaussian embed- ding for clustering uncertain graphs,” Computers and Electrical Engineering, vol. 101, p. 108066, 2022

    2022

  27. [35]

    On embedding uncertain graphs,

    J. Hu, R. Cheng, Z. Huang, Y. Fang, and S. Luo, “On embedding uncertain graphs,” in ACM CIKM , 2017, pp. 157–166

    2017

  28. [36]

    K- nearest neighbors in uncertain graphs,

    M. Potamias, F. Bonchi, A. Gionis, and G. Kollios, “K- nearest neighbors in uncertain graphs,” Proceedings of the VLDB Endowment , vol. 3, no. 1-2, pp. 997–1008, 2010

    2010

  29. [37]

    Core decomposition of uncertain graphs,

    F. Bonchi, F. Gullo, A. Kaltenbrunner, and Y. Volkovich, “Core decomposition of uncertain graphs,” in ACM SIGKDD, 2014, pp. 1316–1325

    2014

  30. [38]

    Shortest paths and centrality in uncer- tain networks,

    A. Saha, R. Brokkelkamp, Y. Velaj, A. Khan, and F. Bonchi, “Shortest paths and centrality in uncer- tain networks,” Proceedings of the VLDB Endowment , vol. 14, no. 7, pp. 1188–1201, 2021

    2021

  31. [39]

    Word representations via gaussian embedding,

    L. Vilnis and A. McCallum, “Word representations via gaussian embedding,” arXiv preprint arXiv:1412.6623, 2014

    2014 arXiv

  32. [40]

    Confidence-based graph convolutional networks for semi-supervised learning,

    S. Vashishth, P. Yadav, M. Bhandari, and P. Taluk- dar, “Confidence-based graph convolutional networks for semi-supervised learning,” in AISTATS, 2019, pp. 1792–1801

    2019

  33. [41]

    Rethinking missing data: Aleatoric uncertainty-aware recommendation,

    C. Wang, F. Feng, Y. Zhang, Q. Wang, X. Hu, and X. He, “Rethinking missing data: Aleatoric uncertainty-aware recommendation,” IEEE Transac- tions on Big Data , 2023

    2023

  34. [42]

    Modeling uncertainty to improve personalized recommendations via bayesian deep learning,

    X. Wang and S. Kadıo˘ glu, “Modeling uncertainty to improve personalized recommendations via bayesian deep learning,” IJDSA, 2023

    2023

  35. [43]

    Uncer- tainty aware semi-supervised learning on graph data,

    X. Zhao, F. Chen, S. Hu, and J.-H. Cho, “Uncer- tainty aware semi-supervised learning on graph data,” NeurIPS, pp. 12 827–12 836, 2020

    2020

  36. [44]

    Uncertainty aware graph gaussian process for semi- supervised learning,

    Z.-Y. Liu, S.-Y. Li, S. Chen, Y. Hu, and S.-J. Huang, “Uncertainty aware graph gaussian process for semi- supervised learning,” in AAAI, 2020, pp. 4957–4964

    2020

  37. [45]

    Bayesian semi- supervised learning with graph gaussian processes,

    Y. C. Ng, N. Colombo, and R. Silva, “Bayesian semi- supervised learning with graph gaussian processes,” NeurIPS, vol. 31, 2018

    2018

  38. [46]

    Ud- gnn: Uncertainty-aware debiased training on semi- homophilous graphs,

    Y. Liu, X. Ao, F. Feng, and Q. He, “Ud- gnn: Uncertainty-aware debiased training on semi- homophilous graphs,” in ACM SIGKDD , 2022, pp. 1131–1140

    2022

  39. [47]

    Accurate and scalable estimation of epistemic uncertainty for graph neural networks,

    P. Trivedi, M. Heimann, R. Anirudh, D. Koutra, and J. J. Thiagarajan, “Accurate and scalable estimation of epistemic uncertainty for graph neural networks,” in ICLR, 2024

    2024

  40. [48]

    A general framework for quantifying aleatoric and epis- temic uncertainty in graph neural networks,

    S. Munikoti, D. Agarwal, L. Das, and B. Natarajan, “A general framework for quantifying aleatoric and epis- temic uncertainty in graph neural networks,” Neuro- computing, vol. 521, pp. 1–10, 2023

    2023

  41. [49]

    Uag: Uncertainty- aware attention graph neural network for defending adversarial attacks,

    B. Feng, Y. Wang, and Y. Ding, “Uag: Uncertainty- aware attention graph neural network for defending adversarial attacks,” in AAAI, 2021, pp. 7404–7412

    2021

  42. [50]

    Un- certainty quantification over graph with conformalized graph neural networks,

    K. Huang, Y. Jin, E. Candes, and J. Leskovec, “Un- certainty quantification over graph with conformalized graph neural networks,” NeurIPS, vol. 36, 2024

    2024

  43. [51]

    Fuzzy neural network for representation learning on uncertain graphs,

    Y.-N. Lin, H.-C. Cai, C.-Y. Zhang, H.-Y. Yao, and C. P. Chen, “Fuzzy neural network for representation learning on uncertain graphs,” IEEE Transactions on Fuzzy Systems, 2024

    2024

  44. [52]

    Fuzzy representation learning on graph,

    C.-Y. Zhang, Y.-N. Lin, C. P. Chen, H.-Y. Yao, H.-C. Cai, and W.-P. Fang, “Fuzzy representation learning on graph,” IEEE Transactions on Fuzzy Systems , vol. 31, no. 10, pp. 3358–3370, 2023

    2023

  45. [53]

    Fuzzy graph subspace convolutional network,

    J. Zhou, Q. Zhang, S. Zeng, and B. Zhang, “Fuzzy graph subspace convolutional network,”IEEE TNNLS, 2022

    2022

  46. [54]

    Uncertainty in graph neural networks: A survey,

    F. Wang, Y. Liu, K. Liu, Y. Wang, S. Medya, and P. S. Yu, “Uncertainty in graph neural networks: A survey,” arXiv preprint arXiv:2403.07185 , 2024

    2024 arXiv

  47. [55]

    Adversarial attack and defense on graph data: A survey,

    L. Sun, Y. Dou, C. Yang, K. Zhang, J. Wang, S. Y. Philip, L. He, and B. Li, “Adversarial attack and defense on graph data: A survey,” IEEE TKDE , pp. 7693–7711, 2022

    2022

  48. [56]

    Adversar- ial graph augmentation to improve graph contrastive learning,

    S. Suresh, P. Li, C. Hao, and J. Neville, “Adversar- ial graph augmentation to improve graph contrastive learning,” NeurIPS, vol. 34, pp. 15 920–15 933, 2021

    2021

  49. [57]

    Towards robust graph contrastive learning,

    N. Jovanovi´ c, Z. Meng, L. Faber, and R. Wattenhofer, “Towards robust graph contrastive learning,” arXiv preprint arXiv:2102.13085, 2021

    2021 arXiv

  50. [58]

    Adversarial defense framework for graph neural network,

    S. Wang, Z. Chen, J. Ni, X. Yu, Z. Li, H. Chen, and P. S. Yu, “Adversarial defense framework for graph neural network,” arXiv preprint arXiv:1905.03679 , 2019

    1905 arXiv

  51. [59]

    Adversarial examples for graph data: deep insights into attack and defense,

    H. Wu, C. Wang, Y. Tyshetskiy, A. Docherty, K. Lu, and L. Zhu, “Adversarial examples for graph data: deep insights into attack and defense,” in IJCAI, 2019, pp. 4816–4823

    2019

  52. [60]

    Exploiting global graph homophily for generalized defense in graph neural networks,

    D. Li, H. Wu, M. Xie, X. Wu, Z. Wu, and W. Zhang, “Exploiting global graph homophily for generalized defense in graph neural networks,” arXiv preprint arXiv:2406.03833, 2024

    2024 arXiv

  53. [61]

    Learning to drop: Robust graph neural network via topological denoising,

    D. Luo, W. Cheng, W. Yu, B. Zong, J. Ni, H. Chen, and X. Zhang, “Learning to drop: Robust graph neural network via topological denoising,” in ACM WSDM , 2021, pp. 779–787

    2021

  54. [62]

    Learning graph embedding with adversarial training methods,

    S. Pan, R. Hu, S.-f. Fung, G. Long, J. Jiang, and C. Zhang, “Learning graph embedding with adversarial training methods,” IEEE Transactions on Cybernetics, pp. 2475–2487, 2019

    2019

  55. [63]

    Defensevgae: Defending against adversarial attacks on graph data via a variational graph autoencoder,

    A. Zhang and J. Ma, “Defensevgae: Defending against adversarial attacks on graph data via a variational graph autoencoder,” arXiv preprint arXiv:2006.08900 , 2020

    2006 arXiv

  56. [64]

    Enhancing the resilience of graph neural networks to topological perturbations in sparse graphs,

    S. He, J. Zhuang, D. Wang, L. Peng, and J. Song, “Enhancing the resilience of graph neural networks to topological perturbations in sparse graphs,” arXiv preprint arXiv:2406.03097, 2024

    2024 arXiv

  57. [65]

    Effi- cient robustness certificates for discrete data: Sparsity- aware randomized smoothing for graphs, images and more,

    A. Bojchevski, J. Gasteiger, and S. G¨ unnemann, “Effi- cient robustness certificates for discrete data: Sparsity- aware randomized smoothing for graphs, images and more,” in ICML, 2020. A Additional Results A.1 Results for Intermediate Perturbation Levels We present results f...

    2020

  58. [66]

    However, REGE consistently outperforms REGE-NCT, highlighting the effectiveness of curricu- lum learning

    Observe that REGE-NCT still demonstrates com- petitive performance, surpassing several state-of-the-art methods. However, REGE consistently outperforms REGE-NCT, highlighting the effectiveness of curricu- lum learning. The results presented are for 10% pertur- bation levels. A...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.