Pith. sign in

REVIEW 4 major objections 5 minor 66 references

A Unified Invariant Learning Framework for Graph Classification

T0 review · 4 major / 5 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 This paper claims that enforcing structural invariance in graph space—matching class-level graphons across environments—alongside semantic invariance lets GNNs recover minimal stable features and generalize out of distribution.

desk verdict Solid empirical recipe for graph OOD with the graphon-distance regularizer, but the proof of the central convergence claim is circular; worth a careful referee, not acceptance as is. read the letter →

arxiv 2501.12595 v1 pith:G6KFA46I submitted 2025-01-22 cs.LG cs.AI

Yongduo Sui , Jie Sun , Shuyao Wang , Zemin Liu , Qing Cui , Longfei Li , Xiang Wang This is my paper · ORCID
classification cs.LGcs.AI
keywords invariantlearninggraphclassificationout-of-distributiongeneralizationneuralnetworksgraphonstructuralinvariancesemanticminimalstablefeatures
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

This paper argues that invariant graph learning has been enforcing invariance in the wrong place: representation-space (semantic) losses alone can mistake 'stable-environmental' subgraphs—features that happen to be stable within the training environments but have no causal link to the label—for true stable features. To fix this, the authors propose Unified Invariant Learning (UIL), which adds a structural invariance principle in graph space: for each label, the stable subgraphs extracted from different environments should be generated by the same graphon, so UIL estimates a stable graphon per class and environment and penalizes their distance. Combined with the usual semantic invariance loss, this joint objective is claimed to drive the extracted stable features toward the minimal stable (causal) features, and the paper gives a theoretical statement (Proposition 2) plus experiments showing better OOD accuracy than existing methods.

What carries the argument

The load-bearing object is the stable graphon: a symmetric measurable function describing, for a class of graphs sharing a label, the probability that an edge of the extracted stable subgraph exists between two nodes. UIL estimates graphons as step-function matrices, aligns graphs by node degrees under the stable edge mask, and enforces structural invariance by minimizing the Frobenius-norm distance between stable graphons across inferred environments; a weak regularity lemma justifies approximation in cut norm. The semantic branch trains the classifier on the extracted stable subgraph, regularizes mask sparsity with a learnable stable ratio, and adds a loss that combines each stable representation with environmental representations so predictions are invariant to environment; environments are inferred from K-means clusters of environmental representations.

What would settle it

Construct a synthetic graph classification task where the causal stable subgraph for each label is drawn from a different graphon in each training environment (labels still fully determined by the subgraph), while environmental subgraphs are identical across environments; if UIL's extracted stable subgraphs still match ground truth better than semantic-only baselines, the assumption survives, but if it collapses to ERM-level accuracy and precision, the graphon-invariance premise is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that invariant graph learning fails when constrained to representation space alone, because features that are stable only within training environments—'stable-environmental features'—can satisfy semantic invariance without being causal. UIL adds a structural invariance principle: for each label, the estimated stable subgraphs across environments should be generated by the same graphon, so the framework estimates a stable graphon per class and environment and penalizes their cut distance (implemented as Frobenius distance between step-function matrices). Under Assumption 1 that minimal stable and estimated stable features are graphon-generated and the graphons are environment-invariant, Proposition 2 states that optimizing the combined objective (sufficiency, structural, and semantic losses) drives the estimated stable features toward minimal stable features indefinitely. Empirically, UIL reports 70.90 average accuracy versus 66.11 for ERM across Motif, CMNIST, Molhiv, and Molbbbp, and higher AUC for recovering ground-truth stable subgraphs than semantic-only baselines.

Load-bearing premise

The argument leans on the assumption that, for every set of graphs sharing a label, both the true causal subgraphs and the extracted ones can be generated by graphons, and that these graphons do not change across environments; if stable causal structure is not graphon-like or environment-specific structural variation exists, the structural loss cannot tell stable features from stable-environmental ones.

Editorial extensions

If this is right

  • If UIL is right, adding a graphon-matching loss to any subgraph-selection GNN can recover causal subgraphs that semantic-invariance-only methods miss.
  • OOD graph classification on synthetic and molecular benchmarks should improve substantially over ERM and over invariant baselines, with the reported gap (70.90 vs 66.11 average) being repeatable.
  • The stable ratio can be learned rather than fixed, because structural invariance supplies the signal that prevents the mask from absorbing stable-environmental features.
  • Graphon distance across environments becomes a diagnostic: smaller cross-environment cut distance of extracted stable subgraphs indicates better recovery of minimal stable features.
  • The unified objective gives a template for combining graph-space and representation-space invariance, applicable beyond classification to any graph task with stable substructures.

Reading between the lines

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

  • A testable extension: apply the structural-invariance loss to node classification or link prediction where stable features are graphon-like; if the same convergence holds, UIL's principle generalizes beyond graph-level tasks.
  • The framework depends on training environments being inferred reliably; when K-means clusters do not align with true environments, the graphon-matching penalty could push together graphons from different causal regimes. One could replace K-means with a differentiable environment estimator and test whether gains persist.
  • The claim suggests a simpler criterion for dataset design: distribution shifts that alter the generative pattern of stable features are out of reach, so benchmarks for invariant graph learning should be constructed with environment-invariant causal graphons, otherwise no invariant method can succeed.
  • Because the structural loss is computed in graph space, the learned stable graphon itself may serve as an interpretability tool: a class-level summary of the causal structure a GNN relies on.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes UIL, a graph-classification method that enforces invariance in both graph space and representation space. A feature separator produces stable and environmental subgraph masks; class-wise stable graphons are estimated from the masked stable features, and the method minimizes Frobenius distances between these graphons across inferred environments (structural invariance), together with sufficiency, regularization, and semantic-invariance losses. The central theoretical claim (Proposition 2) is that, under the full objective in Eq. (8), the estimated stable features converge to the minimal stable feature set G_st. Experiments on SYN, Motif, CMNIST, Molhiv, and Molbbbp report accuracy and AUC gains over ERM and several invariant-learning baselines, with additional analyses of stable-feature precision and cut distances.

Significance. If Proposition 2 were rigorously established, the paper would make a valuable contribution by showing that graphon-level structural invariance complements semantic invariance and helps isolate causal subgraphs for OOD generalization. The framework is clearly described, the code is released, and the experimental comparison is broad, including synthetic and real benchmarks. However, the central convergence proof is invalid as written, and the synthetic experiment used to validate it is generated from the same toy model as the theory. The empirical accuracy improvements in Table 2 are consistent with the method's usefulness, but they do not by themselves establish the paper's mechanistic claim that structural invariance recovers the minimal stable features. The contribution is therefore not yet supported at the level required for publication.

major comments (4)
  1. [Appendix A.2 (proof of Proposition 2)] The proof of Proposition 2 is not valid. It defines the estimated stable graphon as W_hat_st = W_st + W_se ⊙ (1 - W_st), models W_hat_st(i,j) as a Gaussian N(W_st(i,j) + p(1 - W_st(i,j)), eps p), and then asserts that minimizing the Frobenius distance between W_hat^e_st and W_hat^{e'}_st in Eq. (4) propels the value of p towards zero. However, p is the parameter of the Bernoulli distribution of stable-environmental features in the data-generating process; it is not a parameter of f = omega ∘ h ∘ Phi and it does not appear in the objective in Eq. (8). The trainable quantities are the masks M^x and M^a. Minimizing the Frobenius norm of a zero-mean random matrix cannot drive its variance parameter to zero unless that parameter is an optimization variable. The proof therefore assumes the desired conclusion: unless the masks already exclude the environmental edges, the loss does not know about p. Since Proposition 2 is the paper's central convergence claim, this gap is load-bearing.
  2. [Definition 3 and Section 4] The paper equates MinStable(Y), defined in Definition 3 as a minimal subset of features satisfying E_Ptr(Y|G') = E_Ptr(Y|G), with the causal ground-truth G_st. This equivalence is asserted rather than derived. Under the training distribution, stable-environmental features can be predictive of Y, so the minimal sufficient subgraph need not coincide with the causal subgraph; an identifiability argument based on interventions or an explicit causal model is required. Without such an argument, the target of the structural invariance loss is ambiguous, and Proposition 2 cannot be read as a statement about recovering the true causal structure.
  3. [Section 5.2] The synthetic validation in Section 5.2 is generated from exactly the toy model used in the theoretical discussion: SYN-b motifs augmented with Bernoulli(p) stable-environmental subgraphs, with p = 0.001 at training and p = 0.003 at test. Because this is the same data-generating process assumed in the proof of Proposition 2, the experiment cannot provide independent evidence that structural invariance recovers causal subgraphs in general graph distributions. It demonstrates behavior under the paper's own assumptions, not the general convergence claim stated in Section 4.
  4. [Assumption 1 and Definition 5] The convergence claim also rests on Assumption 1, which postulates that the minimal stable features and the estimated stable features are generated by graphons, and on the implicit assumption that the same graphon is shared across environments. These are substantive modeling assumptions about the data-generating process rather than consequences of the optimization. The paper provides no evidence, beyond the synthetic construction, that real-world stable structures such as molecular functional groups satisfy this graphon-invariance condition. Consequently, the scope of Proposition 2 is narrower than the general statement in the abstract and introduction.
minor comments (5)
  1. [Section 2.2] There is a typo 'grouth-truth' in the phrase 'the stable features identified by the semantic invariance might surpass the scope of the grouth-truth stable features'; it should read 'ground-truth'.
  2. [Eq. (7)] The display of Eq. (7) contains malformed bracket notation in the submitted version; please clean up the formatting so the expectation and summation are unambiguous.
  3. [Section 2.1] The citation to CAL in 'causal features [41? ]' appears as a broken or unresolved reference; please correct it.
  4. [Table 4 and Figure 6(a)] Figure 6(a) reports results on SYN-0.1 and SYN-0.9, but Table 4 only lists SYN-b base; please define these datasets and explain how they relate to the SYN-b construction.
  5. [Figure 5] The visualization of captured stable features would be easier to interpret if the ground-truth stable features were marked in every panel, not only in some, and if the caption stated which dataset and environment are shown.

Circularity Check

1 steps flagged · score 7.0 of 10

Proposition 2's proof defines the estimated stable graphon using the data-generation Bernoulli parameter p and then asserts that the graphon loss drives p to zero; since p is not an optimization variable, the convergence claim is assumed rather than derived.

  1. self definitional [Appendix A.2, proof of Proposition 2; uses Lemma 2 and Eq. (4)/(8)]
    "Thus, we can model the estimated graphon as a Gaussian distribution \hat{W}_{st}(i,j) ∼ N(W_{st}(i,j)+p(1−W_{st}(i,j)), εp). ... Considering \hat{W}^{e}_{st}(i,j) − \hat{W}^{e'}_{st}(i,j) ∼ N(0, ε(p+p)), the objective corresponds to minimizing a Gaussian matrix with a mean of 0 and a variance of ε(p+p). The L2 norm of the graphon loss propels the value of p towards zero."

    Lemma 2 fixes p as the Bernoulli parameter of the stable-environmental features in the data-generating process (\hat A_st = A_st ∨ A_se, A_se(i,j) ∼ Bernoulli(p)). The proof then writes \hat W_st(i,j) ∼ N(W_st(i,j)+p(1−W_st(i,j)), εp) and says that minimizing ||\hat W^e_st − \hat W^{e'}_st||_F 'propels the value of p towards zero.' But p does not appear in the trainable objective Eq. (8); the optimization variables are the masks M^x/M^a and the parameters of Φ, h, ω. The structural loss Eq. (4) is evaluated on graphons estimated from masked subgraphs, and nothing in the proof relates the gradients of those masks to p. Minimizing the Frobenius norm of a zero-mean random matrix whose variance is ε(p+p) cannot drive p to zero unless p is itself an optimization variable.

full rationale

Proposition 2 is the central theoretical claim of the paper, but its proof in Appendix A.2 reduces to the assertion that the graphon loss drives the toy model's Bernoulli parameter p to zero. In Lemma 2, p is defined as the data-generation parameter of stable-environmental edges, and \hat W_st = W_st + W_se ⊙ (1−W_st) is built from it; in the proof, the same p is treated as the quantity minimized by Eq. (4). Since p is not an argument of Eq. (8) and no derivation connects the mask parameters to p, the proof assumes that the learned masks already delete the stable-environmental subgraph — exactly the claim to be proven. The Section 5.2 synthetic validation is generated from the same toy example ('we utilize the toy example referenced in our theoretical discussion... supplemented with Bernoulli subgraphs... setting parameter p = 0.001'), so it tests the method under the proof's own construction and does not independently establish the general convergence claim. Separately, the identification of MinStable(Y) with the causal G_st in Section 2.2 is asserted rather than derived, but that is an assumption of the framework rather than a reduction. The UIL algorithm and its empirical comparisons on GOOD/OGB are not circular in themselves; what is circular is the theoretical guarantee. Because that guarantee is the advertised first-principles support for structural invariance, the paper's central derivation is partially circular.

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

The framework rests on a strong structural assumption (graphon adequacy), an unproven equality between training-minimal features and causal features, and the standard semantic invariance axioms from IRM. The main free parameters are the mask ratio, the two loss weights, the number of inferred environments, and the number of graphon partitions. No new physical or conceptual entities are introduced beyond the 'stable-environmental feature' notion, which is a distinction rather than an entity.

free parameters (5)
  • stable ratio rho = learnable; initial values tuned in [0.5, 1.0], best around 0.5-0.7
    Section 3.3 and Figure 7a. Controls the sparsity of node and edge masks; performance depends on its initial value.
  • structural invariance weight alpha = not stated; tuned in {0.1, ..., 1.5}
    Equation (8). Chosen by grid search; authors report low sensitivity but it is still a hyperparameter.
  • semantic invariance weight beta = not stated; tuned in {0.1, ..., 1.5}
    Equation (8). Chosen by grid search; authors report low sensitivity but it is still a hyperparameter.
  • number of inferred environments |E_hat| = tuned in {2, 3, 4, 5, 6}
    Section 3.3 and Figure 7b. K in K-means; performance is stated to be insensitive but it is a hyperparameter.
  • graphon partitions N = not stated in paper
    Used in the step function graphon estimation; the value is not specified in the text.
assumptions (5)
  • ad hoc to paper Stable features G_st and estimated stable features G_hat_st can be generated by graphons W_st and W_hat_st (Assumption 1).
    Section 4, Assumption 1. This is the key structural assumption: class-specific stable features share a generative graphon across environments. If false, the structural invariance loss is meaningless.
  • ad hoc to paper MinStable{Y} equates to the causal ground-truth G_st.
    Section 2.1, after Definition 3. The paper asserts this equality without proof; within-training-distribution minimality does not generally equal causal minimality.
  • domain assumption Semantic invariance principle (sufficiency and invariance conditions of Definition 1).
    Section 2.2, Definition 1. Standard invariant learning assumption inherited from IRM; it assumes an optimal model f* exists and that P(Y|Phi(G)) is equal across environments.
  • domain assumption Graph generation with stable and environmental features, where stable features causally determine Y and environmental features do not.
    Section 2.1. This follows prior work [26, 41, 53] and is the causal premise of all invariant graph learning.
  • standard math Weak regularity lemma for graphons.
    Lemma 1, from Lovasz. Standard result used to justify step-function approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Unified Invariant Learning Framework for Graph Classification." pith.science (2026). https://pith.science/paper/G6KFA46I

@misc{pith2026250112595,
  author       = {Pith},
  title        = {Pith review of: A Unified Invariant Learning Framework for Graph Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G6KFA46I}},
  note         = {Machine review of arXiv:2501.12595}
}
read the original abstract

Invariant learning demonstrates substantial potential for enhancing the generalization of graph neural networks (GNNs) with out-of-distribution (OOD) data. It aims to recognize stable features in graph data for classification, based on the premise that these features causally determine the target label, and their influence is invariant to changes in distribution. Along this line, most studies have attempted to pinpoint these stable features by emphasizing explicit substructures in the graph, such as masked or attentive subgraphs, and primarily enforcing the invariance principle in the semantic space, i.e., graph representations. However, we argue that focusing only on the semantic space may not accurately identify these stable features. To address this, we introduce the Unified Invariant Learning (UIL) framework for graph classification. It provides a unified perspective on invariant graph learning, emphasizing both structural and semantic invariance principles to identify more robust stable features. In the graph space, UIL adheres to the structural invariance principle by reducing the distance between graphons over a set of stable features across different environments. Simultaneously, to confirm semantic invariance, UIL underscores that the acquired graph representations should demonstrate exemplary performance across diverse environments. We present both theoretical and empirical evidence to confirm our method's ability to recognize superior stable features. Moreover, through a series of comprehensive experiments complemented by in-depth analyses, we demonstrate that UIL considerably enhances OOD generalization, surpassing the performance of leading baseline methods. Our codes are available at https://github.com/yongduosui/UIL.

Figures

Figures reproduced from arXiv: 2501.12595 by the authors.

Figure 1
Figure 1. The workflow of several prevalent methods [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overview of the proposed Unified Invariant Learning (UIL) framework. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Precision of extracting the minimal stable features [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visualization results of learned stable graphons. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Visualization results of the captured stable features. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 7
Figure 7. Figure 7: Classification performance over 𝜌 and ⋃︀ℰˆ⋃︀. and semantic invariance in a comprehensive manner, an area that remains largely unexplored. Out-of-Distribution Generalization. Performance degradation is common when training and test data follow disparate distribu￾tions—a…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 51 canonical work pages

  1. [1]

    Kartik Ahuja, Ethan Caballero, Dinghuai Zhang, Jean-Christophe Gagnon-Audet, Yoshua Bengio, Ioannis Mitliagkas, and Irina Rish. 2021. Invariance principle meets information bottleneck for out-of-distribution generalization. NeurIPS 34 (2021), 3438–3450

    work page 2021

  2. [2]

    Martin Arjovsky, Léon Bottou, Ishaan Gulrajani, and David Lopez-Paz. 2019. Invariant risk minimization. arXiv preprint arXiv:1907.02893 (2019)

    arXiv 2019

  3. [3]

    Shiyu Chang, Yang Zhang, Mo Yu, and Tommi Jaakkola. 2020. Invariant rational- ization. In ICML. 1448–1458

    work page 2020

  4. [4]

    Tianlong Chen, Yongduo Sui, Xuxi Chen, Aston Zhang, and Zhangyang Wang

    work page

  5. [5]

    Yongqiang Chen, Yatao Bian, Kaiwen Zhou, Binghui Xie, Bo Han, and James Cheng. 2024. Does invariant graph learning via environment augmentation learn invariance? NeurIPS 36 (2024)

    work page 2024

  6. [6]

    Yongqiang Chen, Yonggang Zhang, Yatao Bian, Han Yang, Kaili Ma, Binghui Xie, Tongliang Liu, Bo Han, and James Cheng. 2022. Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphs. In NeurIPS

    work page 2022

  7. [7]

    Elliot Creager, Jörn-Henrik Jacobsen, and Richard Zemel. 2021. Environment inference for invariant learning. In ICML. 2189–2200

    work page 2021

  8. [8]

    Shaohua Fan, Xiao Wang, Yanhu Mo, Chuan Shi, and Jian Tang. 2022. Debias- ing Graph Neural Networks via Learning Disentangled Causal Substructure. In NeurIPS

    work page 2022

Show all 66 references
  1. [9]

    Shaohua Fan, Xiao Wang, Chuan Shi, Peng Cui, and Bai Wang. 2021. General- izing Graph Neural Networks on Out-Of-Distribution Graphs. arXiv preprint arXiv:2111.10657 (2021)

    2021 arXiv

  2. [10]

    Junfeng Fang, Xinglin Li, Yongduo Sui, Yuan Gao, Guibin Zhang, Kun Wang, Xiang Wang, and Xiangnan He. 2024. Exgc: Bridging efficiency and explainability in graph condensation. In WWW. 721–732

    2024

  3. [11]

    Yuan Gao, Junfeng Fang, Yongduo Sui, Yangyang Li, Xiang Wang, Huamin Feng, and Yongdong Zhang. 2024. Graph Anomaly Detection with Bi-level Optimization. In WWW. 4383–4394

    2024

  4. [12]

    Shurui Gui, Xiner Li, Limei Wang, and Shuiwang Ji. 2022. Good: A graph out-of- distribution benchmark. NeurIPS 35 (2022), 2059–2073

    2022

  5. [13]

    Xiaotian Han, Zhimeng Jiang, Ninghao Liu, and Xia Hu. 2022. G-Mixup: Graph Data Augmentation for Graph Classification. In ICML. 8230–8248

    2022

  6. [14]

    JA Hartigan, MA Wong, et al. 1979. AK-Means Clustering Algorithm. Journal of the Royal Statistical Society Series C 28, 1 (1979), 100–108

    1979

  7. [15]

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open graph benchmark: Datasets for machine learning on graphs. In NeurIPS

    2020

  8. [16]

    Svante Janson. 2010. Graphons, cut norm and distance, couplings and rearrange- ments. arXiv:arXiv:1009.2376

    2010 arXiv

  9. [17]

    Wei Jin, Yao Ma, Xiaorui Liu, Xianfeng Tang, Suhang Wang, and Jiliang Tang

  10. [18]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with Graph Convolutional Networks. In ICLR

    2017

  11. [19]

    Taylor, and Mohamed R

    Boris Knyazev, Graham W. Taylor, and Mohamed R. Amer. 2019. Understanding Attention and Generalization in Graph Neural Networks. In NeurIPS. 4204–4214

    2019

  12. [20]

    Kezhi Kong, Guohao Li, Mucong Ding, Zuxuan Wu, Chen Zhu, Bernard Ghanem, Gavin Taylor, and Tom Goldstein. 2022. Robust Optimization As Data Augmen- tation for Large-Scale Graphs. In CVPR. 60–69

    2022

  13. [21]

    David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Remi Le Priol, and Aaron Courville. 2021. Out-of- distribution generalization via risk extrapolation (rex). In ICML. 5815–5826

    2021

  14. [22]

    Haoyang Li, Xin Wang, Ziwei Zhang, and Wenwu Zhu. 2022. Ood-gnn: Out-of- distribution generalized graph neural network. TKDE (2022)

    2022

  15. [23]

    Haoyang Li, Xin Wang, Ziwei Zhang, and Wenwu Zhu. 2022. Out-of-distribution generalization on graphs: A survey. arXiv preprint arXiv:2202.07987 (2022)

    2022 arXiv

  16. [24]

    Haoyang Li, Ziwei Zhang, Xin Wang, and Wenwu Zhu. 2022. Learning invariant graph representations for out-of-distribution generalization. In NeurIPS

    2022

  17. [25]

    Evan Z Liu, Behzad Haghgoo, Annie S Chen, Aditi Raghunathan, Pang Wei Koh, Shiori Sagawa, Percy Liang, and Chelsea Finn. 2021. Just train twice: Improving group robustness without training group information. In ICML. 6781–6792

    2021

  18. [26]

    Gang Liu, Tong Zhao, Jiaxin Xu, Tengfei Luo, and Meng Jiang. 2022. Graph Rationalization with Environment-based Augmentations. InSIGKDD. 1069–1078

    2022

  19. [27]

    Yang Liu, Xiang Ao, Fuli Feng, Yunshan Ma, Kuan Li, Tat-Seng Chua, and Qing He

  20. [28]

    László Lovász. 2012. Large networks and graph limits . Vol. 60. American Mathe- matical Soc

    2012

  21. [29]

    Dongsheng Luo, Wei Cheng, Dongkuan Xu, Wenchao Yu, Bo Zong, Haifeng Chen, and Xiang Zhang. 2020. Parameterized explainer for graph neural network. In NeurIPS

    2020

  22. [30]

    Yanchen Luo, Sihang Li, Yongduo Sui, Junkang Wu, Jiancan Wu, and Xiang Wang

  23. [31]

    Siqi Miao, Miaoyuan Liu, and Pan Li. 2022. Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanism. In ICML. 15524–15543

    2022

  24. [32]

    Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. 2020. Dropedge: Towards deep graph convolutional networks on node classification. In ICLR

    2020

  25. [33]

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. 2020. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. In ICLR

    2020

  26. [34]

    Zheyan Shen, Peng Cui, Jiashuo Liu, Tong Zhang, Bo Li, and Zhitang Chen. 2020. Stable learning via differentiated variable decorrelation. In SIGKDD. 2185–2193

    2020

  27. [35]

    Zheyan Shen, Jiashuo Liu, Yue He, Xingxuan Zhang, Renzhe Xu, Han Yu, and Peng Cui. 2021. Towards out-of-distribution generalization: A survey. arXiv preprint arXiv:2108.13624 (2021)

    2021 arXiv

  28. [36]

    Yongduo Sui, Tianlong Chen, Pengfei Xia, Shuyao Wang, and Bin Li. 2022. To- wards robust detection and segmentation using vertical and horizontal adversarial training. In IJCNN. 1–8

    2022

  29. [37]

    Yongduo Sui, Wenyu Mao, Shuyao Wang, Xiang Wang, Jiancan Wu, Xiangnan He, and Tat-Seng Chua. 2024. Enhancing Out-of-distribution Generalization on Graphs via Causal Attention Learning. ACM TKDD 18, 5 (2024), 1–24

    2024

  30. [38]

    Yongduo Sui, Caizhi Tang, Zhixuan Chu, Junfeng Fang, Yuan Gao, Qing Cui, Longfei Li, Jun Zhou, and Xiang Wang. 2024. Invariant Graph Learning for Causal Effect Estimation. In WWW. 2552–2562

    2024

  31. [39]

    Yongduo Sui, Shuyao Wang, Jie Sun, Zhiyuan Liu, Qing Cui, Longfei Li, Jun Zhou, Xiang Wang, and Xiangnan He. 2024. A Simple Data Augmentation for Graph Classification: A Perspective of Equivariance and Invariance. ACM TKDD (2024)

    2024

  32. [40]

    Yongduo Sui, Xiang Wang, Tianlong Chen, Meng Wang, Xiangnan He, and Tat- Seng Chua. 2023. Inductive Lottery Ticket Learning for Graph Neural Networks. Journal of Computer Science and Technology (2023)

    2023

  33. [41]

    Yongduo Sui, Xiang Wang, Jiancan Wu, Xiangnan He, and Tat-Seng Chua. 2022. Causal Attention for Interpretable and Generalizable Graph Classification. In SIGKDD. 1696–1705

    2022

  34. [42]

    Yongduo Sui, Qitian Wu, Jiancan Wu, Qing Cui, Longfei Li, Jun Zhou, Xiang Wang, and Xiangnan He. 2023. Unleashing the power of graph data augmentation on covariate distribution shift. NeurIPS 36 (2023)

    2023

  35. [43]

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. In ICLR

    2018

  36. [44]

    Riccardo Volpi, Hongseok Namkoong, Ozan Sener, John C Duchi, Vittorio Murino, and Silvio Savarese. 2018. Generalizing to unseen domains via adversarial data augmentation. NeurIPS 31 (2018)

    2018

  37. [45]

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. 2020. Tent: Fully Test-Time Adaptation by Entropy Minimization. In ICLR

    2020

  38. [46]

    Shuyao Wang, Yongduo Sui, Chao Wang, and Hui Xiong. 2024. Unleashing the Power of Knowledge Graph for Recommendation via Invariant Learning. In WWW. 3745–3755

    2024

  39. [47]

    Shuyao Wang, Yongduo Sui, Jiancan Wu, Zhi Zheng, and Hui Xiong. 2024. Dy- namic Sparse Learning: A Novel Paradigm for Efficient Recommendation. In WSDM. 740–749

    2024

  40. [48]

    Tan Wang, Chang Zhou, Qianru Sun, and Hanwang Zhang. 2021. Causal Atten- tion for Unbiased Visual Recognition. In CVPR. 3091–3100

    2021

  41. [49]

    Yanfang Wang, Yongduo Sui, Xiang Wang, Zhenguang Liu, and Xiangnan He

  42. [50]

    Yiwei Wang, Wei Wang, Yuxuan Liang, Yujun Cai, and Bryan Hooi. 2021. Mixup for node and graph classification. In WWW. 3663–3674

    2021

  43. [51]

    Jiancan Wu, Yi Yang, Yuchun Qian, Yongduo Sui, Xiang Wang, and Xiangnan He

  44. [52]

    Qitian Wu, Hengrui Zhang, Junchi Yan, and David Wipf. 2022. Handling Distri- bution Shifts on Graphs: An Invariance Perspective. In ICLR

    2022

  45. [53]

    Yingxin Wu, Xiang Wang, An Zhang, Xiangnan He, and Tat-Seng Chua. 2022. Discovering Invariant Rationales for Graph Neural Networks. In ICLR

    2022

  46. [54]

    Hongteng Xu, Dixin Luo, Lawrence Carin, and Hongyuan Zha. 2021. Learning graphons via structured gromov-wasserstein barycenters. InAAAI, Vol. 35. 10505– 10513

    2021

  47. [55]

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural Networks?. In ICLR

    2019

  48. [56]

    Nanyang Ye, Kaican Li, Haoyue Bai, Runpeng Yu, Lanqing Hong, Fengwei Zhou, Zhenguo Li, and Jun Zhu. 2022. OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization. In CVPR. 7947–7958

    2022

  49. [57]

    GIF: A general graph unlearning strategy via influence function. In WWW. 651–661

  50. [58]

    Guibin Zhang, Yanwei Yue, Kun Wang, Junfeng Fang, Yongduo Sui, Kai Wang, Yuxuan Liang, Dawei Cheng, Shirui Pan, and Tianlong Chen. 2024. Two Heads Are Better Than One: Boosting Graph Sparse Training via Semantic and Topological Awareness. In ICML. PMLR

    2024

  51. [59]

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. 2018. mixup: Beyond empirical risk minimization. ICLR (2018). A Unified Invariant Learning Framework for Graph Classification KDD ’25, August 3–7, 2025, Toronto, ON, Canada Table 4: Statistics of datasets used...

    2018

  52. [63]

    Zhitao Ying, Dylan Bourgeois, Jiaxuan You, Marinka Zitnik, and Jure Leskovec

  53. [2019]

    In NeurIPS

    GNNExplainer: Generating Explanations for Graph Neural Networks. In NeurIPS. 9240–9251

  54. [2020]

    In SIGKDD

    Graph structure learning for robust graph neural networks. In SIGKDD. 66–74

  55. [2021]

    A unified lottery ticket hypothesis for graph neural networks. In ICML. PMLR, 1695–1706

  56. [2022]

    Inter- national Journal of Intelligent Systems 37, 5 (2022), 3006–3024

    Exploring lottery ticket hypothesis in media recommender systems. Inter- national Journal of Intelligent Systems 37, 5 (2022), 3006–3024

    2022

  57. [2023]

    In SIGKDD

    FLOOD: A Flexible Invariant Learning Framework for Out-of-Distribution Generalization on Graphs. In SIGKDD

  58. [2024]

    Masked Graph Modeling with Multi-View Contrast. ICDE

Pith tools

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