Pith. sign in

REVIEW 5 major objections 6 minor 55 references

Wide & Deep Learning for Node Classification

T0 review · 5 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A graph net that pairs a wide linear branch with a deep GCN reaches new state-of-the-art results in node classification.

desk verdict A competent incremental GCN variant whose full-supervised SOTA claim rests on a retrained GCNII baseline that underperforms the published numbers; worth reviewing, but the headline needs rework. read the letter →

arxiv 2505.02020 v1 pith:RYO6KVG4 submitted 2025-05-04 cs.LG cs.AIstat.ML

Yancheng Chen , Wenguo Yang , Zhipeng Jiang This is my paper · ORCID
classification cs.LGcs.AIstat.ML
keywords nodeclassificationgraphconvolutionalnetworksWide&DeeparchitectureGCNIIIGCNIIover-generalizationinitialresidualfeatures
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that the best way to improve node classification is to stop choosing between shallow and deep graph networks and instead run both at once. It introduces GCNIII, which combines a wide linear classifier whose output can be smoothed by the graph via Intersect memory with a deep GCN built from Initial residual and Identity mapping connections, merging the two with a single weight $\gamma$. On semi-supervised, full-supervised, and inductive benchmarks, GCNIII reports accuracy gains over GCNII on every dataset, with the largest gains on heterophilous web networks. The paper also diagnoses a phenomenon it calls over-generalization—training error persistently above validation error in deep GCNII—and argues that dropout is the key driver, with a theorem showing that beyond a point adding layers barely changes the output.

What carries the argument

The object is the joint prediction $P=\mathrm{Softmax}(\gamma W(X)+(1-\gamma)D(A,X))$, where $W$ is a linear classifier on node features (optionally multiplied by the graph operator $\tilde{G}$ for Intersect memory) and $D$ is a deep GCN with Initial residual (mixing $H^{(0)}$ into every layer with weight $\alpha_l$) and Identity mapping (weight $\beta_l=\lambda/l$, pushing each layer's linear transform toward the identity). The scalar $\gamma$ controls how much the output leans on memorization versus graph-based generalization. The convergence claim is Theorem 4.1: for every $\epsilon>0$ there is a $K_0$ such that for all $K>K_0$, $\|f_{K+1}(A,X)-f_K(A,X)\|_2<\epsilon$, which the paper uses to argue that ultra-deep GCNII is unnecessary.

What would settle it

Fix one GCNIII configuration per dataset by selecting layers, $\gamma$, and the technique set on validation accuracy alone; if the resulting accuracy no longer beats GCNII on the same splits, the paper's unified-architecture state-of-the-art claim is refuted.

Watch

Extended reading notes

Core claim

The central claim is that the Wide & Deep recipe transfers from recommender systems to node classification: a wide, essentially linear branch memorizes raw feature-class associations, a deep GCN branch generalizes through graph propagation, and jointly training them inside one softmax output balances the two. The paper presents GCNIII as the first graph version of this architecture and repackages Initial residual, Identity mapping, dropout, and Intersect memory—a graph-convolution smoothing applied to the wide branch's output—as switchable hyperparameters. Empirically, it reports higher accuracy than GCNII on all ten benchmark datasets plus PPI, and it explains GCNII's behavior through over-generalization, identifying dropout as the key component that keeps deep GCNII training error above validation error.

Load-bearing premise

The central claim stands or falls on whether the per-dataset architecture and hyperparameter choices reported in Appendix B reflect principled model selection rather than tuning to the test set; if they were test-set choices, the comparison against fixed GCNII baselines is not a fair test of one unified model.

Editorial extensions

If this is right

  • Practitioners can treat depth, wide-branch weight, and the three techniques as modular hyperparameters rather than choosing between shallow and deep GCN families.
  • Deep GCNII's effective behavior is concentrated in the layers nearest the output, so ultra-deep 64-layer configurations can be replaced by 8-to-32-layer GCNIII variants with comparable or better accuracy and faster training.
  • Dropout on the input feature embedding is a major source of deep GCN generalization; removing it drops Cora accuracy from above 85% to about 82%.
  • Node feature quality, not only graph structure, drives performance: sparse bag-of-words features markedly outperform random, dense, and one-hot features in the paper's 2-layer GCN experiments.
  • The over-generalization observation identifies a distinct failure mode—training error persistently above validation error—that deep GCN designers should monitor alongside over-smoothing.

Reading between the lines

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

  • If the per-dataset configuration is essential, the paper's real contribution may be a searchable space of techniques plus a diagnostic, rather than one fixed model; a testable extension is to select a single configuration per dataset on validation only and lock it for all datasets.
  • Theorem 4.1 suggests a practical effective-depth rule: stop adding layers once consecutive GCNII outputs are close in spectral norm, which could be checked cheaply on any dataset and compared with validation accuracy.
  • The LLM feature-engineering proposal is untested in this paper, but it implies a concrete experiment: replace bag-of-words features with text-attribute LLM embeddings on the same datasets, keep the wide branch on the sparse features, and see whether the gap over GCNII widens.
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

5 major / 6 minor

Summary. The paper proposes GCNIII, a Wide & Deep architecture for node classification that combines a linear wide branch (optionally with an "Intersect memory" graph-convolution operation) with a deep GCN component built from Initial residual and Identity mapping techniques. The authors introduce the term "over-generalization" for the observed phenomenon where a 64-layer GCNII has higher training error than validation error, present Theorem 4.1 as evidence that ultra-deep GCNII is unnecessary, and evaluate GCNIII on semi-supervised (Cora, Citeseer, Pubmed), full-supervised (those three plus Chameleon, Cornell, Texas, Wisconsin), and inductive (PPI) tasks. The paper also sketches a conceptual extension in which large language models construct node features. Code is released at https://github.com/CYCUCAS/GCNIII.

Significance. If the empirical claims were fully supported, GCNIII would be a simple and useful contribution: it unifies existing GCNII techniques and a linear branch in one framework, the semi-supervised baselines are retrained in a shared framework with 10 seeds, raw per-run numbers are reported in Appendix F, and the code is publicly available. The current manuscript, however, does not yet support the headline "state-of-the-art" claim. The full-supervised comparison in Table 4 reuses published baselines but retrains GCNII under a different framework with explicitly lower results; the GCNIII configuration changes across datasets in Appendix B to the point that on Chameleon the model is no longer a deep Wide & Deep network; the ablation in Table 6 uses a single seed; and the semi-supervised improvements are small relative to the reported standard deviations. Theorem 4.1 is heavily assumption-laden and is not load-bearing for the architecture's empirical value. These issues are fixable within the scope of a major revision, but the paper's central claims currently outrun its evidence.

major comments (5)
  1. [§6.3, Table 4] The headline claim of outperforming GCNII on all seven full-supervised datasets rests on an inconsistent baseline comparison. For GCN, GAT, Geom-GCN variants, and APPNP the paper reuses metrics from Chen et al. (2020), but GCNII is retrained inside the authors' DGL framework, and the text states that on Chameleon, Cornell, Texas, and Wisconsin "the results we get ... are much lower than those reported in Chen et al. (2020)" (e.g., GCNII 54.4 on Chameleon, 59.46 on Cornell, 65.68 on Texas, 65.69 on Wisconsin). The 10-18 point margins over GCNIII on these datasets are therefore an artifact of a weakened baseline. Please report the original GCNII numbers from Chen et al. (2020) alongside the retrained ones, or retrain all baselines under the same framework; otherwise the "state-of-the-art" claim in the abstract and conclusion is not supported.
  2. [Appendix B, Tables 7-9] The GCNIII configuration is changed per dataset in almost every hyperparameter: 2-32 layers, α in [0,0.8], λ in [0,1.5], γ in [0.02,0.1], dropout in [0,0.8], and the technique vector varies. On Chameleon (Table 8), GCNIII uses 2 layers, α=0, λ=0, dropout=0, and techniques [1,0,0], which is essentially a shallow GCN with an Intersect-memory linear branch rather than a deep Wide & Deep model. If these choices were selected by peeking at test performance, the comparison is a per-dataset tuning exercise, not an evaluation of a single proposed architecture. Please describe the model-selection protocol (e.g., selection on validation), or fix a single architecture and tune only standard hyperparameters.
  3. [§6.5, Table 6] The ablation of the three techniques uses a single random seed (the text says "we fixed the random seed as 42, so the results have a certain randomness"). The observed differences are small (-0.4, +1.0, +0.1, +0.6, etc.) compared with the roughly 0.4-0.6 standard deviations reported in Table 2, so the conclusions that Initial residual is "the most influential factor" and the other two techniques "have destabilizing effects" are not supported. Please run the ablation over multiple seeds and report means with standard deviations, ideally with significance tests.
  4. [Theorem 4.1, Appendix A] Theorem 4.1 is not a general statement about GCNII depth because the proof assumes that ReLU is removed, that inputs and intermediate activations are non-negative, that products of Identity-mapping parameter matrices are bounded by a constant C, and, crucially, that the parameters of f_K and f_{K+1} satisfy the "misaligned equality" W_I^{(l)} = tilde W_I^{(l+1)}. This last assumption identifies the extra layer's parameters with an existing layer's parameters, so the difference f_{K+1} - f_K collapses essentially by construction. The theorem should be reframed as a heuristic or proved under substantially weaker and more explicit assumptions; the notation also appears to confuse node count with hidden width (W_I^{(l)} is said to be in R^{n×n}).
  5. [§6.2, Table 2] The semi-supervised SOTA claim rests on differences that are small relative to the reported variability: GCNIII vs GCNII is 85.6±0.4 vs 85.2±0.4 on Cora, 73.0±0.5 vs 72.8±0.6 on Citeseer, and 80.4±0.4 vs 79.8±0.4 on Pubmed. With 10 seeds these gaps are not shown to be statistically significant. Please add significance tests or confidence intervals, and soften the "new state-of-the-art on all three datasets" claim unless the improvements are supported.
minor comments (6)
  1. [Section 4] The "Dropout is the key" claim that removing all dropout from GCNII drops accuracy from "over 85%" to 82% is reported without configuration details or number of seeds; please add experimental details.
  2. [Section 5] "Graph Transoformer" is a typo for "Graph Transformer".
  3. [Appendix F, Table 14 caption] "Mdoel" is a typo for "Model".
  4. [Table 4 and Section 6.5] The column header "Pumb." should be "Pubmed", and "Wide Componet" should be "Wide Component".
  5. [Appendix C, Tables 10-11] The conclusion that Batch Normalization reduces generalization is based on single numbers without seed variability; please report means and standard deviations.
  6. [Section 3.3 and Section 6] Section 6 says LLMs are not used in the experiments, yet the abstract says the paper "explores the use of large language models (LLMs) for node feature engineering"; please clarify that the LLM part is a conceptual proposal, not an evaluated contribution.

Circularity Check

1 steps flagged · score 2.0 of 10

No central circularity; the empirical SOTA claim rests on independent benchmarks. One local theorem-level step is by-construction: the proof of Theorem 4.1 aligns parameters so f_K and f_{K+1} are close by design.

  1. self definitional [Section 4, Theorem 4.1 and Appendix A proof]
    "We use W(l) I and \tilde W(l) I to represent the Identity mapping parameters of the layers in fK(A, X) and fK+1(A, X), respectively. It is important to emphasize that these parameters are misaligned equality, i.e., W(l) I = \tilde W(l+1) I , as our analysis above shows that GCNII is primarily influenced by the layers closer to the output."

    The theorem concludes that for large K, f_{K+1}(A,X) and f_K(A,X) are within epsilon. The proof imposes W_I^{(l)} = \tilde W_I^{(l+1)} and removes ReLU, so the first K layers of f_{K+1} compute exactly the same representations as f_K, up to the final-layer terms that decay as (1-alpha)^K. The closeness of the two models is therefore built into the alignment assumption rather than derived from the GCNII dynamics; a K+1-layer model with these shifted parameters is, by construction, essentially the K-layer model plus a vanishing term. This is a local, non-load-bearing circularity: the theorem is used only to argue that ultra-deep GCNII is unnecessary, and the paper's SOTA claim is supported by experiments rather than by this proof.

full rationale

GCNIII's central claim is empirical: Equations (7)-(12) define a Wide & Deep model, and every accuracy or F1 result in Tables 2-5 is obtained by training against external baselines, so no headline number is algebraically recovered from a fitted parameter or from the model definition itself. The per-dataset configurations in Appendix B (e.g., Chameleon with 2 layers, alpha=0, lambda=0, techniques [1,0,0]) mean that the 'new SOTA' statements are really per-dataset architecture selection; that is a comparison-validity concern, not circularity. Similarly, Table 4 retrains GCNII and reports numbers the authors say are much lower than Chen et al. (2020)'s; this weakens the baseline but does not make the GCNIII outputs equivalent to the comparison by construction. The only step with a by-construction flavor is Theorem 4.1's proof, whose parameter-alignment assumption makes f_K and f_{K+1} close by design. Because that theorem is not load-bearing for the empirical contributions, the overall circularity score is 2 rather than higher.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The model's hyperparameters are extensively dataset-specific, and the proof of Theorem 4.1 relies on two ad hoc assumptions (bounded product C and misaligned equality) that are not independently justified. No new physical or mathematical entities are introduced.

free parameters (6)
  • gamma (wide/deep mixing weight) = 0.01 to 0.1 depending on dataset, Tables 7 and 8
    Chosen by hand per dataset; the paper says adjusting gamma "has strong skills" and that it cannot be trained, so the balance between memorization and generalization is a fitted hyperparameter.
  • alpha_l (initial residual weight) = 0 to 0.8 per dataset, Tables 7-9
    Tuned per dataset; on Chameleon it is 0, which disables initial residual entirely.
  • lambda (identity mapping scaling) = 0 to 1.5 per dataset, Tables 7-9
    Tuned per dataset; on Cora full-supervised lambda is 0, meaning identity mapping is not used.
  • number of layers = 2 to 64 depending on dataset
    Chosen per dataset; heterophily web datasets use 2 or 3 layers, while citation datasets use 8 to 64.
  • techniques vector (Intersect memory, Initial residual, Identity mapping) = per-dataset 0/1 toggles, e.g., [1,0,0] for Chameleon
    Each technique is switched on or off per dataset, so the claimed framework is not a fixed architecture.
  • dropout rate = 0 to 0.8 per dataset
    Tuned per dataset; Chameleon uses dropout 0 while Wisconsin uses 0.8.
assumptions (6)
  • domain assumption Input node features are non-negative and parameters can map non-negative inputs to non-negative outputs, allowing ReLU to be removed in the analysis.
    Adopted from Chen et al. (2020) in Appendix A to make the proof tractable; it is not verified for the datasets.
  • ad hoc to paper The product of any number of identity mapping parameter matrices is bounded above by a constant C.
    Imposed directly in Appendix A as "a stronger assumption" to bound the error terms in Theorem 4.1.
  • ad hoc to paper Identity mapping parameters of f_K and f_{K+1} satisfy the misaligned equality W_I^{(l)} = \tilde W_I^{(l+1)}.
    Assumed in Appendix A so that the main difference between successive depths vanishes; this is close to assuming the theorem's conclusion.
  • domain assumption Weight decay ensures all parameter matrices have bounded l2 norm and bounded products.
    Used in Appendix A to justify a finite bound on weight products; weight decay encourages but does not guarantee the assumed bound in any fixed-width network.
  • standard math The normalized self-loop adjacency matrix G~ has spectral norm at most 1 and I - G~ has spectral norm below 2.
    Follows from Wu et al. (2019) Theorem A.1, cited in Appendix A.
  • domain assumption Fixed train/validation/test splits (Yang et al., 2016 and 60/20/20 splits) are an unbiased evaluation of the models.
    Used throughout Section 6 to compare methods; the paper criticizes other splitting practices but does not show its own splits are more favorable or unbiased.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Wide & Deep Learning for Node Classification." pith.science (2026). https://pith.science/paper/RYO6KVG4

@misc{pith2026250502020,
  author       = {Pith},
  title        = {Pith review of: Wide & Deep Learning for Node Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RYO6KVG4}},
  note         = {Machine review of arXiv:2505.02020}
}
read the original abstract

Wide & Deep, a simple yet effective learning architecture for recommendation systems developed by Google, has had a significant impact in both academia and industry due to its combination of the memorization ability of generalized linear models and the generalization ability of deep models. Graph convolutional networks (GCNs) remain dominant in node classification tasks; however, recent studies have highlighted issues such as heterophily and expressiveness, which focus on graph structure while seemingly neglecting the potential role of node features. In this paper, we propose a flexible framework GCNIII, which leverages the Wide & Deep architecture and incorporates three techniques: Intersect memory, Initial residual and Identity mapping. We provide comprehensive empirical evidence showing that GCNIII can more effectively balance the trade-off between over-fitting and over-generalization on various semi- and full- supervised tasks. Additionally, we explore the use of large language models (LLMs) for node feature engineering to enhance the performance of GCNIII in cross-domain node classification tasks. Our implementation is available at https://github.com/CYCUCAS/GCNIII.

Figures

Figures reproduced from arXiv: 2505.02020 by the authors.

Figure 1
Figure 1. Training error and validation error of the semi-supervised task on Cora with 2-layer vanilla GCN (left) and 64-layer GCNII (right). The training error of deep GCNII is very volatile and much higher than the validation error. We call this phenomenon over￾generalization. networks (Gilmer et al., 2017), such as GCNs, have sev￾eral limitations including homophily assumption (Zhu et al., 2020; Luan et al., 2022) and lack… view at source ↗
Figure 2
Figure 2. Wide & Deep architecture GCNIII. in a single layer. DropEdge (Rong et al., 2020) randomly removes a certain number of edges from the input graph at each training epoch, acting like a data augmenter and also a message passing reducer. DAGNN (Liu et al., 2020) decou￾pling the entanglement of representation transformation and propagation in current graph convolution operations learns graph node representations by adapt… view at source ↗
Figure 3
Figure 3. LLM for GCNIII. Sparse discrete features of graph nodes can be constructed using LLM, such as bag-of-words representa￾tion of document, which can be used in both the Wide and Deep Components. A unified text-attribute description format can also be used to construct text-attribute graphs (TAGs) as input to the LLM, generating dense continuous features that enhance the learn￾ing of the Deep Component. In the field of … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Example network architecture for 64-layer GCNII. The color gradient from black to white represents the weight βl of the Linear Transformation from large to small. Initial residual inputs H(0) directly to each layer, and the network between layers 56 and 64 contains an …
Figure 5
Figure 5. Figure 5: Training error of the semi-supervised task on Cora with GCN, GCNII and GCNIII [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Degree distribution of misclassified nodes of 2-layer GCN and 64-layer GCNII on Cora. Out-of-distribution generalization refers to the model’s ability to maintain strong performance when tested on data that differs from the distribution of the training data. For graph …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 41 canonical work pages

  1. [1]

    Stochastic training of graph convolutional networks with variance reduction

    Chen, J., Zhu, J., and Song, L. Stochastic training of graph convolutional networks with variance reduction. In ICML, 2018

    work page 2018

  2. [2]

    Simple and deep graph convolutional networks

    Chen, M., Wei, Z., Huang, Z., Ding, B., and Li, Y. Simple and deep graph convolutional networks. In ICML, 2020

    work page 2020

  3. [3]

    Exploring the potential of large language models (llms) in learning on graphs

    Chen, Z., Mao, H., Li, H., Jin, W., Wen, H., Wei, X., Wang, S., Yin, D., Fan, W., Liu, H., and Tang, J. Exploring the potential of large language models (llms) in learning on graphs. arXiv preprint arXiv:2307.03393, 2024

    arXiv 2024

  4. [4]

    Wide & deep learning for recommender systems

    Cheng, H.-T., Koc, L., Harmsen, J., Shaked, T., Chandra, T., Aradhye, H., Anderson, G., Corrado, G., Chai, W., Ispir, M., Anil, R., Haque, Z., Hong, L., Jain, V., Liu, X., and Shah, H. Wide & deep learning for recommender systems. In DLRS, 2016

    work page 2016

  5. [5]

    Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks

    Chiang, W.-L., Liu, X., Si, S., Li, Y., Bengio, S., and Hsieh, C.-J. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. In KDD, 2019

    work page 2019

  6. [6]

    Polynormer: Polynomial-expressive graph transformer in linear time

    Deng, C., Yue, Z., and Zhang, Z. Polynormer: Polynomial-expressive graph transformer in linear time. In ICLR, 2024

    work page 2024

  7. [7]

    and Lenssen, J

    Fey, M. and Lenssen, J. E. Fast graph representation learning with pytorch geometric. In ICLR, 2019

    work page 2019

  8. [8]

    Predict then propagate: Graph neural networks meet personalized pagerank

    Gasteiger, J., Bojchevski, A., and Günnemann, S. Predict then propagate: Graph neural networks meet personalized pagerank. In ICLR, 2019

    work page 2019

Show all 55 references
  1. [9]

    S., Riley, P

    Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O., and Dahl, G. E. Neural message passing for quantum chemistry. In ICML, 2017

    2017

  2. [10]

    L., Ying, R., and Leskovec, J

    Hamilton, W. L., Ying, R., and Leskovec, J. Inductive representation learning on large graphs. In ICLR, 2017

    2017

  3. [11]

    Deep residual learning for image recognition

    He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In CVPR, 2016

    2016

  4. [12]

    Harnessing explanations: Llm-to-lm interpreter for enhanced text-attributed graph representation learning

    He, X., Bresson, X., Laurent, T., Perold, A., LeCun, Y., and Hooi, B. Harnessing explanations: Llm-to-lm interpreter for enhanced text-attributed graph representation learning. In ICLR, 2024

    2024

  5. [13]

    and Szegedy, C

    Ioffe, S. and Szegedy, C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 2015

    2015

  6. [14]

    Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. In ICLR, 2015

    2015

  7. [15]

    Kipf, T. N. and Welling, M. Semi-supervised classification with graph convolutional networks. In ICLR, 2017

    2017

  8. [16]

    Deeper insights into graph convolutional networks for semi-supervised learning

    Li, Q., Han, Z., and Wu, X.-M. Deeper insights into graph convolutional networks for semi-supervised learning. In AAAI, 2018

    2018

  9. [17]

    One for all: Towards training one graph model for all classification tasks

    Liu, H., Feng, J., Kong, L., Liang, N., Tao, D., Chen, Y., and Zhang, M. One for all: Towards training one graph model for all classification tasks. In ICLR, 2024 a

    2024

  10. [18]

    Towards deeper graph neural networks

    Liu, M., Gao, H., and Ji, S. Towards deeper graph neural networks. In KDD, 2020

    2020

  11. [19]

    Class-imbalanced graph learning without class rebalancing

    Liu, Z., Qiu, R., Zeng, Z., Yoo, H., Zhou, D., Xu, Z., Zhu, Y., Weldemariam, K., He, J., and Tong, H. Class-imbalanced graph learning without class rebalancing. In ICML, 2024 b

    2024

  12. [20]

    Revisiting heterophily for graph neural networks

    Luan, S., Hua, C., Lu, Q., Zhu, J., Zhao, M., Zhang, S., Chang, X.-W., and Precup, D. Revisiting heterophily for graph neural networks. In ICLR, 2022

    2022

  13. [21]

    Fast graph sharpness-aware minimization for enhancing and accelerating few-shot node classification

    Luo, Y., Chen, Y., Qiu, S., Wang, Y., Zhang, C., Zhou, Y., Cao, X., and Tang, J. Fast graph sharpness-aware minimization for enhancing and accelerating few-shot node classification. In NeurIPS, 2024 a

    2024

  14. [22]

    Classic gnns are strong baselines: Reassessing gnns for node classification

    Luo, Y., Shi, L., and Wu, X.-M. Classic gnns are strong baselines: Reassessing gnns for node classification. In NeurIPS, 2024 b

    2024

  15. [23]

    Is homophily a necessity for graph neural networks? In ICLR, 2022

    Ma, Y., Liu, X., Shah, N., and Tang, J. Is homophily a necessity for graph neural networks? In ICLR, 2022

    2022

  16. [24]

    The pagerank citation ranking: Bringing order to the web

    Page, L., Brin, S., Motwani, R., and Winograd, T. The pagerank citation ranking: Bringing order to the web. In The Web Conference, 1999

    1999

  17. [25]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., Desmaison, A., Köpf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., and Chintala, S. Pytorch: An impera...

    2019

  18. [26]

    C.-C., Lei, Y., and Yang, B

    Pei, H., Wei, B., Chang, K. C.-C., Lei, Y., and Yang, B. Geom-gcn: Geometric graph convolutional networks. In ICLR, 2020

    2020

  19. [27]

    Multi-track message passing: Tackling oversmoothing and oversquashing in graph learning via preventing heterophily mixing

    Pei, H., Li, Y., Deng, H., Hai, J., Wang, P., Ma, J., Tao, J., Xiong, Y., and Guan, X. Multi-track message passing: Tackling oversmoothing and oversquashing in graph learning via preventing heterophily mixing. In ICML, 2024

    2024

  20. [28]

    A critical look at the evaluation of gnns under heterophily: Are we really making progress? In ICLR, 2023

    Platonov, O., Kuznedelev, D., Diskin, M., Babenko, A., and Prokhorenkova, L. A critical look at the evaluation of gnns under heterophily: Are we really making progress? In ICLR, 2023

    2023

  21. [29]

    P., Luu, A

    Rampášek, L., Galkin, M., Dwivedi, V. P., Luu, A. T., Wolf, G., and Beaini, D. Recipe for a general, powerful, scalable graph transformer. In NeurIPS, 2022

    2022

  22. [30]

    Dropedge: Towards deep graph convolutional networks on node classification

    Rong, Y., Huang, W., Xu, T., and Huang, J. Dropedge: Towards deep graph convolutional networks on node classification. In ICLR, 2020

    2020

  23. [31]

    Multi-scale attributed node embedding

    Rozemberczki, B., Allen, C., and Sarkar, R. Multi-scale attributed node embedding. Journal of Complex Networks, 2021

    2021

  24. [32]

    E., Hinton, G

    Rumelhart, D. E., Hinton, G. E., and Williams, R. J. Learning representations by back-propagating errors. Nature, pp.\ 533–536, 1986

    1986

  25. [33]

    Collective classification in network data

    Sen, P., Namata, G., Bilgic, M., Getoor, L., Galligher, B., and EliassiRad, T. Collective classification in network data. AI magazine, pp.\ 29(3), 2008

    2008

  26. [34]

    Ordered gnn: Ordering message passing to deal with heterophily and over-smoothing

    Song, Y., Zhou, C., Wang, X., and Lin, Z. Ordered gnn: Ordering message passing to deal with heterophily and over-smoothing. In ICLR, 2023

    2023

  27. [35]

    Dropout: a simple way to prevent neural networks from overfitting

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., and Salakhutdinov, R. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, pp.\ 1929–1958, 2014

    1929

  28. [36]

    N., Kaiser, L., and Polosukhin, I

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., and Polosukhin, I. Attention is all you need. In NeurIPS, 2017

    2017

  29. [37]

    Graph attention networks

    Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., and Bengio, Y. Graph attention networks. In ICLR, 2018

    2018

  30. [38]

    Deep graph library: A graph-centric, highly-performant package for graph neural networks

    Wang, M., Zheng, D., Ye, Z., Gan, Q., Li, M., Song, X., Zhou, J., Ma, C., Yu, L., Gai, Y., Xiao, T., He, T., Karypis, G., Li, J., and Zhang, Z. Deep graph library: A graph-centric, highly-performant package for graph neural networks. arXiv preprint arXiv:1909.01315, 2020

    1909 arXiv

  31. [39]

    H., Fifty, C., Yu, T., and Weinberger, K

    Wu, F., Zhang, T., de Souza Jr., A. H., Fifty, C., Yu, T., and Weinberger, K. Q. Simplifying graph convolutional networks. In ICML, 2019

    2019

  32. [40]

    Nodeformer: A scalable graph structure learning transformer for node classification

    Wu, Q., Zhao, W., Li, Z., Wipf, D., and Yan, J. Nodeformer: A scalable graph structure learning transformer for node classification. In NeurIPS, 2022

    2022

  33. [41]

    Sgformer: Simplifying and empowering transformers for large-graph representations

    Wu, Q., Zhao, W., Yang, C., Zhang, H., Nie, F., Jiang, H., Bian, Y., and Yan, J. Sgformer: Simplifying and empowering transformers for large-graph representations. In NeurIPS, 2023

    2023

  34. [42]

    Less is more: on the over-globalizing problem in graph transformers

    Xing, Y., Wang, X., Li, Y., Huang, H., and Shi, C. Less is more: on the over-globalizing problem in graph transformers. In ICML, 2024

    2024

  35. [43]

    Representation learning on graphs with jumping knowledge networks

    Xu, K., Li, C., Tian, Y., Sonobe, T., ichi Kawarabayashi, K., and Jegelka, S. Representation learning on graphs with jumping knowledge networks. In ICML, 2018

    2018

  36. [44]

    How powerful are graph neural networks? In ICLR, 2019

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

    2019

  37. [45]

    S., ichi Kawarabayashi, K., and Jegelka, S

    Xu, K., Zhang, M., Li, J., Du, S. S., ichi Kawarabayashi, K., and Jegelka, S. How neural networks extrapolate: From feedforward to graph neural networks. In ICLR, 2021

    2021

  38. [46]

    Graph neural networks are inherently good generalizers: Insights by bridging gnns and mlps

    Yang, C., Wu, Q., Wang, J., and Yan, J. Graph neural networks are inherently good generalizers: Insights by bridging gnns and mlps. In ICLR, 2023

    2023

  39. [47]

    W., and Salakhutdinov, R

    Yang, Z., Cohen, W. W., and Salakhutdinov, R. Revisiting semi-supervised learning with graph embeddings. In ICML, 2016

    2016

  40. [48]

    Do transformers really perform bad for graph representation? In NeurIPS, 2021

    Ying, C., Cai, T., Luo, S., Zheng, S., Ke, G., He, D., Shen, Y., and Liu, T.-Y. Do transformers really perform bad for graph representation? In NeurIPS, 2021

    2021

  41. [49]

    A complete expressiveness hierarchy for subgraph gnns via subgraph weisfeiler-lehman tests

    Zhang, B., Feng, G., Du, Y., He, D., and Wang, L. A complete expressiveness hierarchy for subgraph gnns via subgraph weisfeiler-lehman tests. In ICML, 2023 a

    2023

  42. [50]

    Rethinking the expressive power of gnns via graph biconnectivity

    Zhang, B., Luo, S., Wang, L., and He, D. Rethinking the expressive power of gnns via graph biconnectivity. In ICLR, 2023 b

    2023

  43. [51]

    Evaluating deep graph neural networks

    Zhang, W., Sheng, Z., Jiang, Y., Xia, Y., Gao, J., Yang, Z., and Cui, B. Evaluating deep graph neural networks. arXiv preprint arXiv:2108.00955, 2021

    2021 arXiv

  44. [52]

    Online gnn evaluation under test-time graph distribution shifts

    Zheng, X., Song, D., Wen, Q., Du, B., and Pan, S. Online gnn evaluation under test-time graph distribution shifts. In ICLR, 2024 a

    2024

  45. [53]

    What is missing in homophily? disentangling graph homophily for graph neural networks

    Zheng, Y., Luan, S., and Chen, L. What is missing in homophily? disentangling graph homophily for graph neural networks. In NeurIPS, 2024 b

    2024

  46. [54]

    Beyond homophily in graph neural networks: Current limitations and effective designs

    Zhu, J., Yan, Y., Zhao, L., Heimann, M., Akoglu, L., and Koutra, D. Beyond homophily in graph neural networks: Current limitations and effective designs. In NeurIPS, 2020

    2020

  47. [55]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.