Pith. sign in

REVIEW 3 major objections 4 minor 60 references

Enhancing Distribution and Label Consistency for Graph Out-of-Distribution Generalization

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read DLG claims that unifying augmented and invariant graph generation under two consistency losses improves graph out-of-distribution generalization beyond existing invariant-learning baselines.

desk verdict DLG is a solid empirical integration of known graph-OOD ideas with a new shared edge-mask modifier; the Ld derivation overclaims and the node-level experiments are missing, but the benchmark results deserve a serious referee. read the letter →

arxiv 2501.04102 v1 pith:M7KRIQWL submitted 2025-01-07 cs.LG cs.AI

Song Wang , Xiaodong Yang , Rashidul Islam , Huiyuan Chen , Minghua Xu , Jundong Li , Yiwei Cai This is my paper · ORCID
classification cs.LGcs.AI
keywords graphout-of-distributiongeneralizationdistributionshiftneuralnetworksinvariantsubgraphdataaugmentationconsistencylabeledgemask
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

Graph neural networks lose accuracy when test graphs come from a different distribution than training graphs. The paper claims the standard two-step remedy—first augment training graphs into new environments, then extract invariant subgraphs—is suboptimal because each step can break a consistency: augmented graphs can drift away from plausible data, and extracted subgraphs can lose their link to the label. DLG instead learns edge masks with GNN encoders and differentiable gumbel-softmax sampling to produce both augmented and invariant graphs in one unified pass, and optimizes a distribution-consistency loss Ld and a label-consistency loss Lc. On TPG, DrugOOD, SP-Motif, MNIST-75sp, Graph-SST2, and Molhiv, the paper reports DLG outperforming IRM, V-REx, DIR, GIL, CIGA, and GALA on most datasets.

What carries the argument

The central mechanism is a shared edge-mask modifier: two GNN encoders (GNNv for invariant graphs, GNNa for augmented graphs) produce node representations, an MLP turns pairs of node representations into per-edge sampling probabilities via a sigmoid dot product, and gumbel-softmax makes the edge sampling differentiable. The same modifier yields Gv and Ga, so augmentation and invariant-subgraph extraction are unified as two kinds of edge modification. Ld maximizes a contrastive surrogate for I(Ga; S), pulling the augmented representation toward same-label support graphs and away from different-label graphs and from the original graph; Lc combines the true-label cross-entropy on Gv with a term qv log qa that keeps the class distributions of Gv and Ga close.

What would settle it

Test DLG on a synthetic OOD benchmark with a known invariant subgraph, comparing it to a version whose Ld is replaced by ordinary supervised contrastive learning on original graphs; if accuracies match, the distribution-consistency mechanism is not the source of the gain, and one could also check whether larger Ld values correspond to larger estimated mutual information between generated augmented graphs and held-out same-label graphs.

Watch

Extended reading notes

Core claim

The paper's central claim is that the usual two-step graph OOD recipe fails on consistency, and that explicitly optimizing distribution consistency and label consistency fixes it. Concretely, DLG uses two GNN encoders plus an MLP to compute per-edge sampling probabilities, then applies gumbel-softmax sampling to generate an invariant graph Gv and an augmented graph Ga from the same input graph. The augmented graph is trained with Ld to maximize a contrastive approximation of mutual information with existing same-label training graphs while staying distinct from the original graph, and both generated graphs are trained with Lc to preserve the original label and align their predicted class distributions. The reported experiments show DLG achieving the best accuracy on most of the synthetic and real-world graph classification datasets considered.

Load-bearing premise

The distribution-consistency loss rests on treating 'the probability that an augmented graph was generated from a support graph' as simply 'the two graphs have the same label,' so the contrastive loss is assumed to be a faithful stand-in for the mutual information between augmented and existing graphs.

Editorial extensions

If this is right

  • If DLG is right, graph OOD generalization can be improved without explicitly partitioning training environments; one modifier plus two consistency losses suffices.
  • Augmented graphs that stay close to existing same-label graphs should reduce unrealistic molecular edits, which matters for drug-affinity prediction on DrugOOD and Molhiv.
  • Label consistency gives a concrete way to keep supervision information in invariant subgraphs, and the ablations suggest this matters most under large distribution shifts such as SP-Motif and MNIST-75sp.
  • The framework's gains on low-dimensional node-feature datasets indicate that consistency-preserving augmentation can retain informativeness where richer features are unavailable.

Reading between the lines

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

  • In the paper's derivation of Ld, p(Ga|Gi) is replaced by the label indicator 1(Y=yi), so the objective is effectively a supervised contrastive loss; the paper does not demonstrate that this surrogate bounds or tracks the true mutual information, so the distribution-consistency claim is only as strong as that leap.
  • Because Ld is computed from the original graph's representation ha rather than from Ga itself, the encoder's stability during training is an unstated requirement: if the encoder changes, the alignment target shifts, and the loss may not reflect augmented-graph consistency.
  • A natural extension the paper only gestures at is applying the same two-consistency modifier to node-level OOD settings; the conclusion claims node-level experiments, but the body reports only graph-level tables, so that transfer is not yet evidenced.
  • The diversity term inside Ld, which pushes augmented representations away from Gv, could be tested directly: measure whether augmented graphs actually explore new label-consistent regions or merely act as a regularizer that improves the encoder.
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

3 major / 4 minor

Summary. The paper proposes DLG, a framework for graph out-of-distribution (OOD) generalization. DLG uses a shared edge-mask modifier to generate both an invariant subgraph Gv and an augmented graph Ga from an input graph G. Two consistency losses are introduced: Ld, intended to maximize the mutual information between the augmented graph and a support set of training graphs (distribution consistency), and Lc, which combines a classification loss on Gv with a self-distillation term between the predictions of Gv and Ga (label consistency). The authors evaluate DLG on TPG, DrugOOD, SP-Motif, MNIST-75sp, Graph-SST2, and Molhiv, reporting improvements over ERM, IRM, V-REx, DIR, GIL, CIGA, and GALA on most datasets.

Significance. The overarching idea of unifying augmentation and invariant subgraph extraction with explicit consistency regularization is timely and could be useful for graph OOD generalization. The label consistency loss Lc in Eq. (10) is a reasonable classification-plus-distillation objective that directly supervises the augmented graph through qa = f(Ga). The code is publicly available, which is a strength. However, the paper's central theoretical justification for Ld—that it maximizes I(Ga; S)—is not supported by the derivation in Eqs. (3)–(9). As written, Ld is a supervised contrastive loss on the representation of the original graph G, not on the augmented graph Ga. This significantly weakens the claimed mechanism, though the empirical gains may still hold if Ld acts as a useful representation regularizer.

major comments (3)
  1. [IV-B, Eqs. (3)-(9)] The derivation of Ld as a mutual information objective is invalid for three reasons. First, Eq. (3) expresses I(Ga; S) as a sum of per-graph terms p(Ga, Gi), which is not the mutual information between Ga and the set S; the correct object would involve the joint distribution of the entire support set. Second, Eq. (6) replaces p(Ga|Gi) with the label indicator 1(Y = yi), which conflates label equality with the conditional data-generation distribution and is not an approximation with a clear error bound. Third, and most importantly, Eq. (9) uses ha, the graph-level representation of the original graph G (as stated in the text: 'as Ga is generated from G, here we use the graph representation of G as ha'), rather than a representation of the sampled augmented graph Ga. Consequently, the gradient of Ld with respect to the sampled adjacency matrix Aa is zero; the loss only regularizes the encoder output Ha, which indirectly affects the mask generator. The ablation 'w/o Ld' therefore removes a representation-level contrastive regularizer, not a distribution-consistency constraint on Ga. The authors should either reframe Ld as an auxiliary representation regularizer or provide a derivation that actually involves the augmented graph.
  2. [V and Conclusion] The abstract and conclusion claim experiments on 'both graph-level and node-level' OOD generalization datasets, but Section V and Table I contain only graph classification results. No node-level datasets, protocols, or results are presented anywhere in the manuscript. This discrepancy should be corrected either by adding node-level experiments or by removing the claim.
  3. [Table I] The dataset aggregation is underspecified. The text states that TPG has four variants and DrugOOD includes six datasets, yet Table I reports only one TPG column and two DrugOOD columns (EC50 and Ki). It is unclear whether the reported numbers are averages over variants/splits, which splits of DrugOOD are used (Assay, Scaffold, Size), and how the standard deviations are computed. Without this information, the reader cannot assess the statistical significance of the reported improvements.
minor comments (4)
  1. [Eq. (8)] The denominator uses an undefined symbol 'hd' in the softmax expression; it should be 'ha'.
  2. [References] Reference [20] lists the journal as 'IEEE TPMI'; the correct abbreviation is 'IEEE TPAMI'.
  3. [Section V-A] The description of Molhiv does not specify the scaffold split or the evaluation protocol used; please provide the exact split and metric.
  4. [Section V-C] The ablation study in Fig. 2 is described only briefly; please include the exact α values tested and the datasets shown in the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the two consistency losses are direct training objectives and the empirical comparison is against external benchmarks.

full rationale

The derivation chain in Section IV-B is not circular in the sense this pass targets. Although the mutual-information derivation is formally loose (Eq. (9) computes Ld from the original graph's representation ha rather than from the sampled augmented graph Ga, so Ld does not literally constrain Ga), that is an approximation or support gap, not a reduction of a prediction to its input by construction. Ld is a concrete supervised-contrastive regularizer on ha, and Lc in Eq. (10) directly enforces label consistency through classification and self-distillation; neither term is defined in terms of the empirical claim being tested. The central empirical claim in Section V-B is supported by comparisons on public benchmarks (TPG, DrugOOD, SP-Motif, MNIST-75sp, Graph-SST2, Molhiv) rather than by a fitted parameter renamed as a prediction. The paper's self-citations (e.g., refs. [6], [13], [17], [29]) are contextual background and are not load-bearing for the main mechanism. Therefore no circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The framework relies on three domain assumptions borrowed from the invariant rationalization literature plus one ad hoc approximation in the distribution consistency loss. The free parameters are the balancing weight alpha and the gumbel-softmax temperature, whose values are not reported. No new physical or conceptual entities are introduced.

free parameters (2)
  • alpha = not reported
    Hyperparameter in Eq. (11) controlling the trade-off between Ld and Lc. Tuned per dataset as implied by the ablation discussion, but the chosen values are not listed.
  • gumbel_softmax_temperature = not reported
    Temperature used in the gumbel-softmax edge sampling described in Section IV-A; no value or schedule is given.
assumptions (4)
  • domain assumption OOD generalization for graphs can be achieved by learning invariant subgraphs and training a classifier on them.
    This is the standard invariant rationalization assumption inherited from DIR and CIGA, used throughout Sections IV-A and IV-C.
  • ad hoc to paper The mutual information between the augmented graph and the support set can be approximated using label equality and the source graph's representation.
    Eqs. (6)-(9) replace p(Ga|Gi) with 1(Y=yi) and use ha from G instead of an embedding of Ga; this makes the MI objective tractable but changes its meaning.
  • domain assumption The classifier trained on the invariant graph transfers to test distributions.
    Used in the label consistency loss Lc; no formal guarantee of invariance across environments is provided.
  • domain assumption Augmented graphs share the label of the source graph.
    The setup states the augmented graph Ga is classified with the same label as G, used in Eq. (6-10), and is a design choice that may not hold under real label shift.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Distribution and Label Consistency for Graph Out-of-Distribution Generalization." pith.science (2026). https://pith.science/paper/M7KRIQWL

@misc{pith2026250104102,
  author       = {Pith},
  title        = {Pith review of: Enhancing Distribution and Label Consistency for Graph Out-of-Distribution Generalization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M7KRIQWL}},
  note         = {Machine review of arXiv:2501.04102}
}
read the original abstract

To deal with distribution shifts in graph data, various graph out-of-distribution (OOD) generalization techniques have been recently proposed. These methods often employ a two-step strategy that first creates augmented environments and subsequently identifies invariant subgraphs to improve generalizability. Nevertheless, this approach could be suboptimal from the perspective of consistency. First, the process of augmenting environments by altering the graphs while preserving labels may lead to graphs that are not realistic or meaningfully related to the origin distribution, thus lacking distribution consistency. Second, the extracted subgraphs are obtained from directly modifying graphs, and may not necessarily maintain a consistent predictive relationship with their labels, thereby impacting label consistency. In response to these challenges, we introduce an innovative approach that aims to enhance these two types of consistency for graph OOD generalization. We propose a modifier to obtain both augmented and invariant graphs in a unified manner. With the augmented graphs, we enrich the training data without compromising the integrity of label-graph relationships. The label consistency enhancement in our framework further preserves the supervision information in the invariant graph. We conduct extensive experiments on real-world datasets to demonstrate the superiority of our framework over other state-of-the-art baselines.

Figures

Figures reproduced from arXiv: 2501.04102 by the authors.

Figure 1
Figure 1. The overall framework of DLG. Given a graph [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The performance of our framework DLG with different [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 56 canonical work pages

  1. [1]

    Arnetminer: extraction and mining of academic social networks,

    J. Tang, J. Zhang, L. Yao, J. Li, L. Zhang, and Z. Su, “Arnetminer: extraction and mining of academic social networks,” in SIGKDD, 2008. 1

    work page 2008

  2. [2]

    Inferring networks of substi- tutable and complementary products,

    J. McAuley, R. Pandey, and J. Leskovec, “Inferring networks of substi- tutable and complementary products,” in SIGKDD, 2015. 1

    work page 2015

  3. [3]

    Meta- gnn: On few-shot node classification in graph meta-learning,

    F. Zhou, C. Cao, K. Zhang, G. Trajcevski, T. Zhong, and J. Geng, “Meta- gnn: On few-shot node classification in graph meta-learning,” in CIKM,

    work page

  4. [4]

    Deep gaussian embedding of graphs: Unsupervised inductive learning via ranking,

    A. Bojchevski and S. G ¨unnemann, “Deep gaussian embedding of graphs: Unsupervised inductive learning via ranking,” in ICLR, 2018. 1

    work page 2018

  5. [5]

    Graph few-shot class-incremental learning,

    Z. Tan, K. Ding, R. Guo, and H. Liu, “Graph few-shot class-incremental learning,” in WSDM, 2022. 1

    work page 2022

  6. [6]

    Contrastive meta-learning for few- shot node classification,

    S. Wang, Z. Tan, H. Liu, and J. Li, “Contrastive meta-learning for few- shot node classification,” in SIGKDD, 2023. 1

    work page 2023

  7. [7]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” in ICLR, 2017. 1

    work page 2017

  8. [8]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Bengio, “Graph attention networks,” in ICLR, 2018. 1

    work page 2018

Show all 60 references
  1. [9]

    Graph neural networks: A review of methods and applications,

    J. Zhou, G. Cui, S. Hu, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, and M. Sun, “Graph neural networks: A review of methods and applications,” AI Open, 2020. 1

    2020

  2. [10]

    Heterogeneous network embedding via deep architectures,

    S. Chang, W. Han, J. Tang, G.-J. Qi, C. C. Aggarwal, and T. S. Huang, “Heterogeneous network embedding via deep architectures,” in SIGKDD, 2015. 1

    2015

  3. [11]

    Inductive representation learning on large graphs,

    W. L. Hamilton, R. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” in NeurIPS, 2017. 1

    2017

  4. [12]

    How powerful are graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” in ICLR, 2019. 1

    2019

  5. [13]

    Graph few-shot learning with task-specific structures,

    S. Wang, C. Chen, and J. Li, “Graph few-shot learning with task-specific structures,” in NeurIPS, 2022. 1

    2022

  6. [14]

    Deep neural networks for learning graph representations,

    S. Cao, W. Lu, and Q. Xu, “Deep neural networks for learning graph representations,” in AAAI, 2016. 1

    2016

  7. [15]

    Interpretable and generalizable graph learning via stochastic attention mechanism,

    S. Miao, M. Liu, and P. Li, “Interpretable and generalizable graph learning via stochastic attention mechanism,” in ICML, 2022. 1

    2022

  8. [16]

    Mind the label shift of augmentation-based graph ood generalization,

    J. Yu, J. Liang, and R. He, “Mind the label shift of augmentation-based graph ood generalization,” in CVPR, 2023. 1

    2023

  9. [17]

    Safety in graph machine learning: Threats and safeguards,

    S. Wang, Y . Dong, B. Zhang, Z. Chen, X. Fu, Y . He, C. Shen, C. Zhang, N. V . Chawla, and J. Li, “Safety in graph machine learning: Threats and safeguards,” arXiv:2405.11034, 2024. 1

    2024 arXiv

  10. [18]

    Handling distribution shifts on graphs: An invariance perspective,

    Q. Wu, H. Zhang, J. Yan, and D. Wipf, “Handling distribution shifts on graphs: An invariance perspective,” in ICLR, 2022. 1, 2, 5

    2022

  11. [19]

    Graphrnn: Generating realistic graphs with deep auto-regressive models,

    J. You, R. Ying, X. Ren, W. Hamilton, and J. Leskovec, “Graphrnn: Generating realistic graphs with deep auto-regressive models,” in ICML,

  12. [20]

    Graph neural networks with convolutional arma filters,

    F. M. Bianchi, D. Grattarola, L. Livi, and C. Alippi, “Graph neural networks with convolutional arma filters,” IEEE TPMI, 2021. 1

    2021

  13. [21]

    Domain adaptation: Learning bounds and algorithms,

    Y . Mansour, M. Mohri, and A. Rostamizadeh, “Domain adaptation: Learning bounds and algorithms,” in COLT, 2009. 1

    2009

  14. [22]

    Generalizing from several related classification tasks to a new unlabeled sample,

    G. Blanchard, G. Lee, and C. Scott, “Generalizing from several related classification tasks to a new unlabeled sample,” in NeurIPS, 2011. 1

    2011

  15. [23]

    Domain generalization via invariant feature representation,

    K. Muandet, D. Balduzzi, and B. Sch ¨olkopf, “Domain generalization via invariant feature representation,” in ICML, 2013. 1

    2013

  16. [24]

    Recognition in terra incognita,

    S. Beery, G. Van Horn, and P. Perona, “Recognition in terra incognita,” in ECCV, 2018. 1

    2018

  17. [25]

    Do imagenet classifiers generalize to imagenet?

    B. Recht, R. Roelofs, L. Schmidt, and V . Shankar, “Do imagenet classifiers generalize to imagenet?” in ICML, 2019. 1

    2019

  18. [26]

    One pixel attack for fooling deep neural networks,

    J. Su, D. V . Vargas, and K. Sakurai, “One pixel attack for fooling deep neural networks,” IEEE TEVC, 2019. 1

    2019

  19. [27]

    Collective spammer detection in evolving multi-relational social networks,

    S. Fakhraei, J. Foulds, M. Shashanka, and L. Getoor, “Collective spammer detection in evolving multi-relational social networks,” in SIGKDD, 2015. 1

    2015

  20. [28]

    Good: A graph out-of-distribution benchmark,

    S. Gui, X. Li, L. Wang, and S. Ji, “Good: A graph out-of-distribution benchmark,” in NeurIPS Datasets and Benchmarks Track , 2022. 1

    2022

  21. [29]

    Few-shot node classification with extremely weak supervision,

    S. Wang, Y . Dong, K. Ding, C. Chen, and J. Li, “Few-shot node classification with extremely weak supervision,” in WSDM, 2023. 1

    2023

  22. [30]

    Out-of-distribution generalization via risk extrapolation (rex),

    D. Krueger, E. Caballero, J.-H. Jacobsen, A. Zhang, J. Binas, D. Zhang, R. Le Priol, and A. Courville, “Out-of-distribution generalization via risk extrapolation (rex),” in ICML, 2021. 1, 4

    2021

  23. [31]

    Invariant rationalization,

    S. Chang, Y . Zhang, M. Yu, and T. Jaakkola, “Invariant rationalization,” in ICML, 2020. 1

    2020

  24. [32]

    Deep learning for segmentation of brain tumors: Impact of cross-institutional training and testing,

    E. A. AlBadawy, A. Saha, and M. A. Mazurowski, “Deep learning for segmentation of brain tumors: Impact of cross-institutional training and testing,” Medical Physics, 2018. 1

    2018

  25. [33]

    Dark model adaptation: Semantic image segmentation from daytime to nighttime,

    D. Dai and L. Van Gool, “Dark model adaptation: Semantic image segmentation from daytime to nighttime,” in ITSC, 2018. 1

    2018

  26. [34]

    Discovering invariant rationales for graph neural networks,

    Y . Wu, X. Wang, A. Zhang, X. He, and T.-S. Chua, “Discovering invariant rationales for graph neural networks,” in ICLR, 2022. 1, 2, 4

    2022

  27. [35]

    Invariant risk minimization,

    M. Arjovsky, L. Bottou, I. Gulrajani, and D. Lopez-Paz, “Invariant risk minimization,” arXiv, 2019. 2, 4

    2019

  28. [36]

    Unsupervised domain adaptation by backpropagation,

    Y . Ganin and V . Lempitsky, “Unsupervised domain adaptation by backpropagation,” in ICML, 2015. 2

    2015

  29. [37]

    Domain generalization with adversarial feature learning,

    H. Li, S. J. Pan, S. Wang, and A. C. Kot, “Domain generalization with adversarial feature learning,” in CVPR, 2018. 2

    2018

  30. [38]

    Graph prototypical networks for few-shot learning on attributed networks,

    K. Ding, J. Wang, J. Li, K. Shu, C. Liu, and H. Liu, “Graph prototypical networks for few-shot learning on attributed networks,” in CIKM, 2020. 2

    2020

  31. [39]

    Return of frustratingly easy domain adaptation,

    B. Sun, J. Feng, and K. Saenko, “Return of frustratingly easy domain adaptation,” in AAAI, 2016. 2

    2016

  32. [40]

    Invariance, causality and robustness,

    P. B ¨uhlmann, “Invariance, causality and robustness,” 2020. 2

    2020

  33. [41]

    Invariant causal prediction for nonlinear models,

    C. Heinze-Deml, J. Peters, and N. Meinshausen, “Invariant causal prediction for nonlinear models,” JCI, 2018. 2

    2018

  34. [42]

    Does distributionally robust supervised learning give robust classifiers?

    W. Hu, G. Niu, I. Sato, and M. Sugiyama, “Does distributionally robust supervised learning give robust classifiers?” in ICML, 2018. 2

    2018

  35. [43]

    Robust optimization over multiple domains,

    Q. Qian, S. Zhu, J. Tang, R. Jin, B. Sun, and H. Li, “Robust optimization over multiple domains,” in AAAI, 2019. 2

    2019

  36. [44]

    Distributionally robust neural networks,

    S. Sagawa, P. W. Koh, T. B. Hashimoto, and P. Liang, “Distributionally robust neural networks,” in ICLR, 2020. 2

    2020

  37. [45]

    Adversarial weight perturbation improves generalization in graph neural networks,

    Y . Wu, A. Bojchevski, and H. Huang, “Adversarial weight perturbation improves generalization in graph neural networks,” in AAAI, 2023. 2

    2023

  38. [46]

    Adversarial causal augmentation for graph covariate shift,

    Y . Sui, X. Wang, J. Wu, A. Zhang, and X. He, “Adversarial causal augmentation for graph covariate shift,” arXiv:2211.02843, 2022. 2

    2022 arXiv

  39. [47]

    Learning invariant graph representations for out-of-distribution generalization,

    H. Li, Z. Zhang, X. Wang, and W. Zhu, “Learning invariant graph representations for out-of-distribution generalization,” in NeurIPS, 2022. 2, 4

    2022

  40. [48]

    Learning causally invariant representations for out-of- distribution generalization on graphs,

    Y . Chen, Y . Zhang, Y . Bian, H. Yang, M. Kaili, B. Xie, T. Liu, B. Han, and J. Cheng, “Learning causally invariant representations for out-of- distribution generalization on graphs,” NeurIPS, 2022. 2, 4

    2022

  41. [49]

    Debiasing graph neural networks via learning disentangled causal substructure,

    S. Fan, X. Wang, Y . Mo, C. Shi, and J. Tang, “Debiasing graph neural networks via learning disentangled causal substructure,” NeurIPS, 2022. 2

    2022

  42. [50]

    Does in- variant graph learning via environment augmentation learn invariance?

    Y . Chen, Y . Bian, K. Zhou, B. Xie, B. Han, and J. Cheng, “Does in- variant graph learning via environment augmentation learn invariance?” Advances in Neural Information Processing Systems , vol. 36, 2023. 2, 4, 5

    2023

  43. [51]

    Categorical reparameterization with gumbel-softmax,

    E. Jang, S. Gu, and B. Poole, “Categorical reparameterization with gumbel-softmax,” in ICLR, 2017. 3

    2017

  44. [52]

    Causal attention for interpretable and generalizable graph classification,

    Y . Sui, X. Wang, J. Wu, M. Lin, X. He, and T.-S. Chua, “Causal attention for interpretable and generalizable graph classification,” in SIGKDD,

  45. [53]

    Representation learning with contrastive predictive coding,

    A. v. d. Oord, Y . Li, and O. Vinyals, “Representation learning with contrastive predictive coding,” arXiv:1807.03748, 2018. 3

    2018 arXiv

  46. [54]

    Parameterized explainer for graph neural network,

    D. Luo, W. Cheng, D. Xu, W. Yu, B. Zong, H. Chen, and X. Zhang, “Parameterized explainer for graph neural network,” Advances in neural information processing systems , vol. 33, pp. 19 620–19 631, 2020. 4

    2020

  47. [55]

    Drugood: Out-of-distribution (ood) dataset curator and benchmark for ai-aided drug discovery–a focus on affinity prediction problems with noise annotations,

    Y . Ji, L. Zhang, J. Wu, B. Wu, L.-K. Huang, T. Xu, Y . Rong, L. Li, J. Ren, D. Xue et al., “Drugood: Out-of-distribution (ood) dataset curator and benchmark for ai-aided drug discovery–a focus on affinity prediction problems with noise annotations,” arXiv:2201.09637, 2022. 4

    2022 arXiv

  48. [56]

    Gnnex- plainer: Generating explanations for graph neural networks,

    Z. Ying, D. Bourgeois, J. You, M. Zitnik, and J. Leskovec, “Gnnex- plainer: Generating explanations for graph neural networks,” in NeurIPS,

  49. [57]

    Understanding attention and generalization in graph neural networks,

    B. Knyazev, G. W. Taylor, and M. Amer, “Understanding attention and generalization in graph neural networks,” in NeurIPS, 2019. 4

    2019

  50. [58]

    Moleculenet: a benchmark for molecular machine learning,

    Z. Wu, B. Ramsundar, E. N. Feinberg, J. Gomes, C. Geniesse, A. S. Pappu, K. Leswing, and V . Pande, “Moleculenet: a benchmark for molecular machine learning,” Chemical Science, 2018. 4

    2018

  51. [59]

    Open graph benchmark: Datasets for machine learning on graphs,

    W. Hu, M. Fey, M. Zitnik, Y . Dong, H. Ren, B. Liu, M. Catasta, and J. Leskovec, “Open graph benchmark: Datasets for machine learning on graphs,” in NeurIPS, 2020. 4

    2020

  52. [60]

    Ogb-lsc: A large-scale challenge for machine learning on graphs,

    W. Hu, M. Fey, H. Ren, M. Nakata, Y . Dong, and J. Leskovec, “Ogb-lsc: A large-scale challenge for machine learning on graphs,” arXiv:2103.09430, 2021. 4

    2021 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.