Pith. sign in

REVIEW 3 major objections 4 minor 46 references

Towards Effective Graph Rationalization via Boosting Environment Diversity

T0 review · 3 major / 4 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 GRBE augments graphs by mixing environment subgraphs in the original graph space, and reports that this produces more diverse training distributions, better out-of-distribution classification, and more accurate rationale subgraphs.

desk verdict Original-space environment mixup is a real idea, but the paper's label-preservation assumption for augmented graphs is unverified and the evidence doesn't support the claimed superiority. read the letter →

arxiv 2412.12880 v1 pith:XWZGW66M submitted 2024-12-17 cs.LG

Yujie Wang , Kui Yu , Yuhong Zhang , Fuyuan Cao , Jiye Liang This is my paper · ORCID
classification cs.LG
keywords graphrationalizationout-of-distributiongeneralizationneuralnetworksdataaugmentationenvironmentdiversitymixupcontrastivelearningsubgraphextraction
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 rationalization splits each input graph into a label-determining rationale subgraph and a label-independent environment subgraph, then augments the environment so that training becomes robust to distribution shift. This paper argues that prior methods augment in representation space, where the pool of environments is small and unevenly sampled, so the augmented distribution is not diverse enough. GRBE instead mixes environment subgraphs directly in the original graph space: it samples edges from two different environments, splices the resulting subgraph onto a rationale, and labels the new graph with the rationale's label. A contrastive refinement step first sharpens the rationale/environment split so that the new graphs remain valid. The paper reports that this raises classification accuracy and rationale-recovery AUC across benchmark datasets, with the largest gains when the train/test shift is severe.

What carries the argument

The method couples two modules: PRSE (Precise Rationale Subgraph Extraction) samples rationale edges from a Bernoulli distribution parameterized by a learned mask, uses a Gumbel-Softmax/Concrete relaxation to keep the sampling differentiable, and refines the mask with self-supervised contrastive pairs (perturb the environment while keeping the rationale for positive pairs, perturb the rationale while keeping the environment for negative pairs). EDA (Environment Diversity Augmentation) takes two environment subgraphs, builds a block-diagonal adjacency $A_{\mathrm{ext}}$ with mask $M_{\mathrm{mix},e}=\mathrm{diag}(\lambda M_{i,e},(1-\lambda)M_{j,e})$, samples a new environment subgraph from that mixed distribution, and splices it onto a rationale subgraph with $N_{\mathrm{add}}$ bridge edges. This machinery moves augmentation from representation space into the original graph space, which is what generates the larger and more varied set of augmented environments.

What would settle it

Run GRBE on Spurious-Motifs where ground-truth motif masks are known, then inspect the augmented graphs: if a non-trivial fraction of mixed graphs inherit a label that differs from the label determined by their ground-truth motif, the label-inheritance assumption is violated and the reported gains cannot be credited to clean environment diversity.

Watch

Extended reading notes

Core claim

The paper's central claim is that environment diversity, not merely environment replacement, is the bottleneck for graph rationalization under distribution shift. Existing methods such as GREA recombine rationale and environment representations, so augmentation can only reshuffle existing environment categories; GRBE treats edges as Bernoulli samples and mixes two environment subgraphs in input space, producing far more environment categories (for example, 34 versus 5 on Spmotif-0.9) and correspondingly better out-of-distribution classification and rationale extraction. The authors report average improvements of 7.65% in rationalization performance and 6.11% in classification performance over comparison methods, with Spmotif classification gains ranging from +9.0173% to +31.951%.

Load-bearing premise

The load-bearing premise is that the learned rationale subgraph alone determines the graph label, so every mixed graph can be assigned the label of the graph whose rationale was reused; if the rationale mask misses or adds even one edge, the augmented samples become mislabeled.

Editorial extensions

If this is right

  • On the Spurious-Motifs datasets, GRBE improves classification accuracy over comparison methods by +9.0173% to +31.951%, and the benefit grows as the train/test distribution shift becomes larger.
  • Rationale extraction quality improves substantially: on Spmotif, rationale-recovery AUC rises by up to +17.556% over the best baseline, and on MUTAG GRBE reaches 99.86% AUC.
  • The environment diversity measure increases across all datasets: GRBE yields 34 to 323 environment-subgraph clusters versus 2 to 157 for the representation-space baseline GREA.
  • Ablation studies show that removing either the contrastive refinement loss or the environment-augmentation loss degrades both classification and rationalization, indicating that the two modules reinforce each other.
  • On datasets with small distribution shifts, such as OGBG-Molbace and OGBG-Molbbbp, GRBE is slightly inferior to baselines, suggesting its advantage is specifically tied to the magnitude of distribution shift.

Reading between the lines

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

  • Editorial: If environment diversity is the active ingredient, sampling $\lambda$ adaptively (for example, biasing toward rare or dissimilar environment clusters) could push diversity further; the paper keeps $\lambda$ as a fixed hyperparameter.
  • Editorial: Original-space environment mixup should transfer to node-level or link-level tasks wherever a rationale subgraph can be identified; the paper validates only graph classification.
  • Editorial: The label-inheritance rule hides a testable risk: if a mixed environment accidentally contains motif-like structure, the augmented sample may be mislabeled. An oracle experiment using ground-truth rationale masks, with augmentation otherwise unchanged, would isolate how much of the gain comes from environment diversity rather than from label noise.
  • Editorial: The clustering-based diversity count could be replaced by a quantitative effective-number-of-environments measure and used as a criterion for selecting which augmented samples to keep, a direction the paper leaves as future work.
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 manuscript proposes GRBE, a graph rationalization method that performs data augmentation in the original graph space rather than in representation space. The method has two modules: PRSE, which refines rationale subgraph extraction through contrastive losses, and EDA, which mixes environment subgraphs from different graphs and attaches the resulting mixed environment to the rationale subgraph, assigning the original graph's label. The authors report experiments on three synthetic Spmotif variants and six real-world datasets, claiming average improvements of 7.65% in rationalization AUC and 6.11% in classification accuracy over state-of-the-art baselines.

Significance. The core idea of boosting environment diversity in the original graph space is a timely and plausible alternative to representation-space augmentation, and the paper provides a concrete implementation with comparisons on nine datasets, ablations, and visualizations. If the label-preservation assumption behind the augmented samples were verified, GRBE would be a useful contribution to graph rationalization. However, the headline claim is weakened by negative results on two datasets and by the lack of direct evidence that augmented graphs are correctly labeled. The significance of the contribution is therefore contingent on additional analysis and validation.

major comments (3)
  1. [IV-B, Eq. (12)] The label assignment of augmented graphs is load-bearing but unverified. Equation (12) sets the label of Gi,aug = Gi,r + Gmix,e to yi, justified only by Definition 1. However, the rationale mask is learned through stochastic Bernoulli sampling (Eqs. 2-3) with a soft sparsity penalty (Eq. 4), and the paper itself notes in Section I that a one-edge difference changes the Spmotif label. Table V reports rationalization AUC of 85.04 on Spmotif-0.9, so a non-negligible fraction of learned masks deviates from ground truth. For those graphs, the augmented sample can correspond to a different ground-truth label than yi, and training on La in Eq. (17) can push the classifier away from invariance. The authors should report label accuracy on augmented samples, add a confidence-based filtering mechanism, or provide a robustness analysis under mask error.
  2. [Abstract and Table IV] The headline average improvements of 7.65% and 6.11% are unweighted means of per-dataset relative improvements and are dominated by the large Spmotif gains. Table IV shows that GRBE is worse than GREA on OGBG-Molbace (-5.56%) and OGBG-Molbbbp (-0.94%), and on MUTAG and Twitter the reported gains are within one standard deviation of the best baseline. The explanation in Section V-C that performance gains track the reported distribution-shift distance is not supported by Table IV: Graph-SST5 has a smaller distance than MUTAG but a larger improvement, while OGBG-Molbbbp has a distance similar to Graph-SST5 yet a negative improvement. The paper should report per-dataset significance tests, use a clearly defined aggregate metric rather than an unweighted mean of relative improvements, and provide a more rigorous account of the negative cases.
  3. [IV-C, Eq. (16)] The contrastive objective in Eq. (9) combines -I(h1,+, h2,+) + I(h, h-), and Eq. (16) is presented as the InfoNCE estimator for I(h', h'') for arbitrary pairs. Using the same estimator for the negative pair (h, h-) and then minimizing it is not a standard negative-pair contrastive objective: minimizing the log-ratio can be achieved by increasing similarity to other negative samples in the denominator, so the intended repulsion between h and h- is not guaranteed. The authors should state the exact loss used in the implementation, including how positive and negative pairs are batched, and provide the derivation or a reference for the negative-pair term.
minor comments (4)
  1. [I, Fig. 2] The text states that GRBE 'can produce 34 categories of environment subgraphs... which is significantly more than the environment subgraphs learned by GRBE'; this appears to be a typo and should read 'learned by GREA', since the comparison is with GREA and Table VI lists 34 categories for GRBE.
  2. [Table I] The notation rows for Gi,e and Gj,e are duplicated with conflicting descriptions, one saying 'rationale subgraphs' and the other saying 'environment subgraphs'; these rows need to be corrected.
  3. [IV-A, Eq. (4)] The sparsity loss is written without an explicit norm, so it is unclear whether the absolute value or an L1/L2 norm is intended; please define the norm explicitly.
  4. [V-F, Fig. 6] The figure plots a 'Distance' between augmented and original distributions, but Table IV also uses 'Distance' to denote training-testing distribution shift. Please clarify whether the same JS divergence is used and whether higher values in Fig. 6 are intended to indicate greater diversity.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity: GRBE's OOD and rationale-discovery claims rest on external benchmark metrics; the only self-citation ([10], co-authors K. Yu and F. Cao) is background context in the introduction and is not load-bearing.

full rationale

The paper is self-contained against external benchmarks, so no circular step is present; the score of 2 reflects the single self-citation, which is not load-bearing. All model parameters (GNN1, MLP1, MLP2) are learned end-to-end from LGRBE = Lr + αLa + βLc + γLs (Eq. 17) on training and augmented data; the reported test ACC and rationale AUC are external metrics that never enter the optimization, so no fitted parameter is renamed as a prediction. The augmented-graph label rule in Eq. 12 (Gi,aug = Gi,r + Gmix,e labeled yi) is an assumption imported from Definition 1's sufficiency and independence conditions rather than a derived result; the paper itself flags the sensitivity of this assumption in Section I ('house' and 'cycle' differ by one edge) and reports no measurement of label accuracy on augmented samples, which is a missing-support limitation for the effectiveness claim. This is weighed as a correctness risk, not a logical circularity, because the claim is externally testable and does not reduce to its inputs by construction. The diversity evidence (cluster counts in Table VI, JS distance in Fig. 6) measures a property EDA was designed to maximize, but the OOD claim is evaluated on held-out test distributions and is conditional: GRBE loses on OGBG-Molbace/BBBP despite higher diversity (Section V-C), so 'diversity to generalization' is not tautological. The only self-citation is [10] (co-authors K. Yu, F. Cao), cited among general OOD methods [3]-[11] in the introduction; no derivation rests on it. No uniqueness theorem, ansatz-by-citation, or renamed-known-result pattern is present.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The method adds no new physical or conceptual entities. Its free parameters are the usual loss weights, augmentation ratios, sparsity levels, and a few unspecified procedural constants (lambda, radd, temperature). The axioms are the standard rationalization assumptions plus standard reparameterization and InfoNCE machinery.

free parameters (9)
  • Loss weight alpha (Lr vs La balance) = 0.5, 1, 0.1, 0.01 per dataset (Table III)
    Balances the original rationale prediction loss against the augmented-graph prediction loss; selected per dataset.
  • Loss weight beta (contrastive loss Lc) = 0.1 for most datasets, 0.01 or 1 elsewhere (Table III)
    Weights the contrastive refinement objective in the total loss.
  • Loss weight gamma (sparsity loss Ls) = 0.1 to 1 per dataset (Table III)
    Weights the soft sparsity constraint on the rationale mask.
  • Augmentation ratio raug = 0.2 to 0.5 per dataset (Table III)
    Fraction of training graphs augmented per epoch; controls the amount of synthetic data.
  • Sparsity level rs = 0.5 to 0.7 per dataset (Table III)
    Targets the average fraction of edges sampled into the rationale subgraph; manually predefined without a principled choice.
  • Mixup coefficient lambda = not reported
    Controls the mixture of the two environment subgraphs in Eq. 10; no default value or sampling distribution is stated.
  • Edge-add ratio radd = 10% (stated in Section IV-B)
    Determines the number Nadd of edges added to connect rationale and mixed environment subgraphs; set without sensitivity analysis.
  • Contrastive dropout probability = 0.5
    Each edge in the perturbed subgraph is dropped with probability 0.5 when constructing positive and negative pairs (Section IV-A).
  • Concrete relaxation temperature t = not reported
    Temperature in Eq. 3 for differentiable sampling; no value is given.
assumptions (5)
  • domain assumption Each graph decomposes into rationale and environment subgraphs satisfying complementarity, sufficiency, and conditional independence (Definition 1).
    Section III, Definition 1. This is the rationalization assumption inherited from prior work (DIR, GREA); the method does not verify it empirically.
  • domain assumption The label of an augmented graph equals the label of the source graph because the rationale subgraph is preserved.
    Section IV-B, Eq. 12. If rationale extraction is imperfect, augmented labels may be wrong.
  • standard math Bernoulli sampling with the concrete/Gumbel-Softmax relaxation is a valid differentiable surrogate for discrete edge selection.
    Eqs. 3 and 11, adapted from Jang et al. [32].
  • standard math InfoNCE is an appropriate estimator of mutual information for the contrastive terms in Eq. 9.
    Eq. 16, from Chen et al. [38]; standard practice in contrastive learning.
  • domain assumption Node/edge dropout of the environment subgraph preserves the label-relevant information of the graph representation.
    Section IV-A; positive pair construction assumes that environment perturbation does not change the semantic content captured by GNN1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Effective Graph Rationalization via Boosting Environment Diversity." pith.science (2026). https://pith.science/paper/XWZGW66M

@misc{pith2026241212880,
  author       = {Pith},
  title        = {Pith review of: Towards Effective Graph Rationalization via Boosting Environment Diversity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XWZGW66M}},
  note         = {Machine review of arXiv:2412.12880}
}
read the original abstract

Graph Neural Networks (GNNs) perform effectively when training and testing graphs are drawn from the same distribution, but struggle to generalize well in the face of distribution shifts. To address this issue, existing mainstreaming graph rationalization methods first identify rationale and environment subgraphs from input graphs, and then diversify training distributions by augmenting the environment subgraphs. However, these methods merely combine the learned rationale subgraphs with environment subgraphs in the representation space to produce augmentation samples, failing to produce sufficiently diverse distributions. Thus, in this paper, we propose to achieve an effective Graph Rationalization by Boosting Environmental diversity, a GRBE approach that generates the augmented samples in the original graph space to improve the diversity of the environment subgraph. Firstly, to ensure the effectiveness of augmentation samples, we propose a precise rationale subgraph extraction strategy in GRBE to refine the rationale subgraph learning process in the original graph space. Secondly, to ensure the diversity of augmented samples, we propose an environment diversity augmentation strategy in GRBE that mixes the environment subgraphs of different graphs in the original graph space and then combines the new environment subgraphs with rationale subgraphs to generate augmented graphs. The average improvements of 7.65% and 6.11% in rationalization and classification performance on benchmark datasets demonstrate the superiority of GRBE over state-of-the-art approaches.

Figures

Figures reproduced from arXiv: 2412.12880 by the authors.

Figure 1
Figure 1. An example to illustrate the difference between the existing general augmentation in the representation space and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The unsupervised clustering results of the environment [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overall framework of our proposed GRBE method. It performs environment diversity augmentation within the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Ablation studies. We report the mean - 0.5*std of the [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Hyper-parameter sensitive analysis. We report the mean - std of classification and rationalization performance under [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: We can see that, in the initial stages, GRBE exhibits a [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The unsupervised clustering results of the environment [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the rationale subgraphs extracted by GRBE on the Spmotif-0.9 and Graph-SST2 datasets. The ground [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 41 canonical work pages

  1. [1]

    Pre-training molecular graph representation with 3d geometry,

    S. Liu, H. Wang, W. Liu, J. Lasenby, H. Guo, and J. Tang, “Pre-training molecular graph representation with 3d geometry,” in Proceedings of the 10th International Conference on Learning Representations , 2022. [Online]. Available: https://openreview.net/forum?id=xQUe1pOKPam

    work page 2022

  2. [2]

    Out-of-distribution generaliza- tion on graphs: A survey,

    H. Li, X. Wang, Z. Zhang, and W. Zhu, “Out-of-distribution generaliza- tion on graphs: A survey,” arXiv preprint arXiv:2202.07987 , 2022

    arXiv 2022

  3. [3]

    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 Proceedings of the 39th International Conference on Machine Learning , 2022, pp. 15 524– 15 543

    work page 2022

  4. [4]

    Individual and structural graph information bottlenecks for out- of-distribution generalization,

    L. Yang, J. Zheng, H. Wang, Z. Liu, Z. Huang, S. Hong, W. Zhang, and B. Cui, “Individual and structural graph information bottlenecks for out- of-distribution generalization,” IEEE Transactions on Knowledge and Data Engineering, pp. 682–693, 2023

    work page 2023

  5. [5]

    Graph invariant learning with subgraph co-mixup for out-of-distribution generalization,

    T. Jia, H. Li, C. Yang, T. Tao, and C. Shi, “Graph invariant learning with subgraph co-mixup for out-of-distribution generalization,” in Pro- ceedings of the 38th AAAI Conference on Artificial Intelligence , 2024, pp. 8562–8570

    work page 2024

  6. [6]

    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,” in Proceedings of the 36th Conference on Neural Information Processing Systems , 2022, pp. 22 131–22 148

    work page 2022

  7. [7]

    Joint learning of label and environment causal independence for graph out-of-distribution general- ization,

    S. Gui, M. Liu, X. Li, Y . Luo, and S. Ji, “Joint learning of label and environment causal independence for graph out-of-distribution general- ization,” in Proceedings of the 38th Conference on Neural Information Processing Systems, 2024, pp. 3945–3978

    work page 2024

  8. [8]

    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 Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022, pp. 1696–1705

    2022

Show all 46 references
  1. [9]

    Ood-gnn: Out-of-distribution generalized graph neural network,

    H. Li, X. Wang, Z. Zhang, and W. Zhu, “Ood-gnn: Out-of-distribution generalized graph neural network,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 7, pp. 7328–7340, 2022

    2022

  2. [10]

    Learning causal representations for robust domain adaptation,

    S. Yang, K. Yu, F. Cao, L. Liu, H. Wang, and J. Li, “Learning causal representations for robust domain adaptation,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 3, pp. 2750–2764, 2021

    2021

  3. [11]

    Dive: Subgraph disagreement for graph out-of-distribution generalization,

    X. Sun, L. Wang, Q. Liu, S. Wu, Z. Wang, and L. Wang, “Dive: Subgraph disagreement for graph out-of-distribution generalization,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 2794–2805

    2024

  4. [12]

    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 Proceedings of the 10th International Conference on Learning Representations , 2022. [Online]. Available: https://openreview.net/forum?id=hGXij5rfiHw

    2022

  5. [13]

    Graph rationalization with environment-based augmentations,

    G. Liu, T. Zhao, J. Xu, T. Luo, and M. Jiang, “Graph rationalization with environment-based augmentations,” in Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2022, pp. 1069–1078

    2022

  6. [14]

    Adversarial causal aug- mentation for graph covariate shift,

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

    2022 arXiv

  7. [15]

    Mean shift: A robust approach toward feature space analysis,

    D. Comaniciu and P. Meer, “Mean shift: A robust approach toward feature space analysis,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 5, pp. 603–619, 2002

    2002

  8. [16]

    Gn- nexplainer: Generating explanations for graph neural networks,

    Z. Ying, D. Bourgeois, J. You, M. Zitnik, and J. Leskovec, “Gn- nexplainer: Generating explanations for graph neural networks,” in Proceedings of the 33rd Conference on Neural Information Processing Systems, 2019, pp. 9244–9255

    2019

  9. [17]

    Does invariant graph learning via environment augmentation learn invariance?

    Y . Chen, Y . Bian, K. Zhou, B. Xie, B. Han, and J. Cheng, “Does invariant graph learning via environment augmentation learn invariance?” vol. 36, 2024

    2024

  10. [18]

    Graph information bottleneck for subgraph recognition,

    J. Yu, T. Xu, Y . Rong, Y . Bian, J. Huang, and R. He, “Graph information bottleneck for subgraph recognition,” in Proceedings of the 9th International Conference on Learning Representations , 2021. [Online]. Available: https://openreview.net/forum?id=bM4Iqfg8M2k JOURNAL OF LA...

    2021

  11. [19]

    Improving subgraph recognition with variational graph information bottleneck,

    J. Yu, J. Cao, and R. He, “Improving subgraph recognition with variational graph information bottleneck,” in Proceedings of the 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 19 396–19 405

    2022

  12. [20]

    Individual and structural graph information bottlenecks for out- of-distribution generalization,

    L. Yang, J. Zheng, H. Wang, Z. Liu, Z. Huang, S. Hong, W. Zhang, and B. Cui, “Individual and structural graph information bottlenecks for out- of-distribution generalization,” IEEE Transactions on Knowledge and Data Engineering, vol. 36, no. 2, pp. 682–693, 2024

    2024

  13. [21]

    Invariant risk minimization,

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

    1907 arXiv

  14. [22]

    Invariance principle meets information bottleneck for out-of-distribution generalization,

    K. Ahuja, E. Caballero, D. Zhang, J.-C. Gagnon-Audet, Y . Bengio, I. Mitliagkas, and I. Rish, “Invariance principle meets information bottleneck for out-of-distribution generalization,” in Proceedings of the 35th Conference on Neural Information Processing Systems , 2021, pp. ...

    2021

  15. [23]

    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 Proceedings of the 38th International Conference on Machine Learning , 2021, pp. 5815–5826

    2021

  16. [24]

    Learning substructure invariance for out-of-distribution molecular representations,

    N. Yang, K. Zeng, Q. Wu, X. Jia, and J. Yan, “Learning substructure invariance for out-of-distribution molecular representations,” Advances in Neural Information Processing Systems , vol. 35, pp. 12 964–12 978, 2022

    2022

  17. [25]

    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 Proceedings of the 36th Conference on Neural Information Processing Systems , 2022, pp. 11 828–11 841

    2022

  18. [26]

    Flood: A flexible invariant learning framework for out-of-distribution generalization on graphs,

    Y . Liu, X. Ao, F. Feng, Y . Ma, K. Li, T.-S. Chua, and Q. He, “Flood: A flexible invariant learning framework for out-of-distribution generalization on graphs,” in Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 1548– 1558

    2023

  19. [27]

    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,” Advances in Neural Information Processing Systems , vol. 35, pp. 24 934–24 946, 2022

    2022

  20. [28]

    Advancing molecule invariant representation via privileged substructure identification,

    R. Wang, H. Dai, C. Yang, L. Song, and C. Shi, “Advancing molecule invariant representation via privileged substructure identification,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 3188–3199

    2024

  21. [29]

    Learning from shortcut: A shortcut-guided approach for graph rationalization,

    L. Yue, Q. Liu, Y . Liu, W. Gao, and C. Song, “Learning from shortcut: A shortcut-guided approach for graph rationalization,” 2024. [Online]. Available: https://openreview.net/forum?id=XcwHDoKvVg

    2024

  22. [30]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 11 620–11 630

    2023

  23. [31]

    Cooperative classi- fication and rationalization for graph generalization,

    L. Yue, Q. Liu, Y . Liu, W. Gao, F. Yao, and W. Li, “Cooperative classi- fication and rationalization for graph generalization,” in Proceedings of the ACM on Web Conference 2024 , 2024, pp. 344–352

    2024

  24. [32]

    Categorical reparameterization with gumbel-softmax,

    E. Jang, S. Gu, and B. Poole, “Categorical reparameterization with gumbel-softmax,” in Proceedings of the 5th International Conference on Learning Representations , 2017. [Online]. Available: https://openreview.net/forum?id=rkE3y85ee

    2017

  25. [33]

    Boosting graph contrastive learning via graph contrastive saliency,

    C. Wei, Y . Wang, B. Bai, K. Ni, D. Brady, and L. Fang, “Boosting graph contrastive learning via graph contrastive saliency,” inProceedings of the 40th International Conference on Machine Learning , 2023, pp. 36 839– 36 855

    2023

  26. [34]

    Adversarial graph augmentation to improve graph contrastive learning,

    S. Suresh, P. Li, C. Hao, and J. Neville, “Adversarial graph augmentation to improve graph contrastive learning,” in Proceedings of the 35th Con- ference on Neural Information Processing Systems , 2021, pp. 15 920– 15 933

    2021

  27. [35]

    Data augmentation for graph neural networks,

    T. Zhao, Y . Liu, L. Neves, O. Woodford, M. Jiang, and N. Shah, “Data augmentation for graph neural networks,” in Proceedings of the 35th AAAI conference on artificial intelligence , 2021, pp. 11 015–11 023

    2021

  28. [36]

    Graph transplant: Node saliency-guided graph mixup with local structure preservation,

    J. Park, H. Shim, and E. Yang, “Graph transplant: Node saliency-guided graph mixup with local structure preservation,” in Proceedings of the 36th AAAI Conference on Artificial Intelligence , 2022, pp. 7966–7974

    2022

  29. [37]

    Do generated data always help contrastive learning?

    Y . Wang, J. Zhang, and Y . Wang, “Do generated data always help contrastive learning?” in Proceedings of the 12th International Conference on Learning Representations , 2024. [Online]. Available: https://openreview.net/forum?id=S5EqslEHnz

    2024

  30. [38]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in Proceedings of the 37th International Conference on Machine Learning , 2020, pp. 1597– 1607

    2020

  31. [39]

    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 Proceedings of the 34th Conference on Neural Information Processing Systems, 2020, pp. 22 118–22 133

    2020

  32. [40]

    Recursive deep models for semantic compositionality over a sentiment treebank,

    R. Socher, A. Perelygin, J. Wu, J. Chuang, C. D. Manning, A. Y . Ng, and C. Potts, “Recursive deep models for semantic compositionality over a sentiment treebank,” in Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing , 2013, pp. 1631– 1642....

    2013

  33. [41]

    Fast and accurate modeling of molecular atomization energies with machine learning,

    M. Rupp, A. Tkatchenko, K.-R. M ¨uller, and O. A. V on Lilienfeld, “Fast and accurate modeling of molecular atomization energies with machine learning,” Physical review letters, vol. 108, no. 5, p. 058301, 2012

    2012

  34. [42]

    How powerful are graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” in Proceedings of the 7th International Conference on Learning Representations , 2019. [Online]. Available: https://openreview.net/forum?id=ryGs6iA5Km

    2019

  35. [43]

    Environment inference for invariant learning,

    E. Creager, J.-H. Jacobsen, and R. Zemel, “Environment inference for invariant learning,” in Proceedings of the 38th International Conference on Machine Learning , 2021, pp. 2189–2200

    2021

  36. [44]

    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,” in Proceedings of the 34th Conference on Neural Information Processing Systems , 2020, pp. 19 620–19 631

    2020

  37. [45]

    Interpreting graph neural networks for nlp with differentiable edge masking,

    M. S. Schlichtkrull, N. D. Cao, and I. Titov, “Interpreting graph neural networks for nlp with differentiable edge masking,” in Proceedings of the 9th International Conference on Learning Representations , 2021. [Online]. Available: https://openreview.net/forum?id=WznmQa42ZAx

    2021

  38. [46]

    The jensen-shannon divergence,

    M. L. Men ´endez, J. Pardo, L. Pardo, and M. Pardo, “The jensen-shannon divergence,” Journal of the Franklin Institute , vol. 334, no. 2, pp. 307– 318, 1997

    1997

Pith tools

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