arxiv: 2602.17071 · v2 · submitted 2026-02-19 · 💻 cs.LG · cs.AI

Recognition: 2 theorem links

· Lean Theorem

AdvSynGNN: Structure-Adaptive Graph Neural Nets via Adversarial Synthesis and Self-Corrective Propagation

Rong Fu , Muge Qi , Chunlei Meng , Shuo Yin , Kun Liu , Zhaolu Kang , Simon Fong

Authors on Pith no claims yet

Pith reviewed 2026-05-15 21:28 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords graph neural networksadversarial learningheterophilystructure adaptationnode classificationcontrastive learningresidual correction

0 comments

The pith

AdvSynGNN combines adversarial structure synthesis with residual label correction to improve graph neural network accuracy on noisy and heterophilous graphs.

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

Graph neural networks lose accuracy when graphs contain structural noise or non-homophilous connections. The paper introduces AdvSynGNN to counter this by generating multi-resolution structures through contrastive objectives for better initial representations. A transformer backbone then adjusts attention using learned topological signals to accommodate heterophily. An adversarial propagation engine proposes connectivity changes while a discriminator maintains overall consistency, and a per-node confidence metric drives residual corrections to refine labels iteratively.

Core claim

AdvSynGNN establishes a framework that uses multi-resolution structural synthesis and contrastive objectives for geometry-sensitive initializations, a transformer backbone modulated by topological signals to handle heterophily, an adversarial propagation engine in which a generative component identifies potential connectivity alterations and a discriminator enforces global coherence, and a residual correction scheme guided by per-node confidence metrics to achieve stable iterative label refinement, resulting in optimized predictive accuracy across diverse graph distributions while preserving computational efficiency.

What carries the argument

The adversarial propagation engine, in which a generative component identifies potential connectivity alterations while a discriminator enforces global coherence, paired with a residual correction scheme guided by per-node confidence metrics.

If this is right

Higher node classification accuracy holds across graphs with varying levels of structural noise.
Adaptive attention in the transformer backbone improves handling of non-homophilous topologies.
The overall architecture maintains computational efficiency suitable for large-scale graphs.
Implementation protocols support reliable deployment without additional preprocessing steps.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The correction scheme could extend to dynamic graphs where edges change over time.
Similar adversarial and residual mechanisms might reduce sensitivity to input noise in other representation learning models.
Real-world applications could require less manual graph cleaning before training.

Load-bearing premise

The adversarial propagation engine and residual correction scheme integrate stably without introducing new instabilities or requiring extensive hyperparameter tuning.

What would settle it

A direct comparison on a standard heterophilous graph benchmark showing no accuracy gain over baseline graph neural networks or a measurable increase in training time.

Figures

Figures reproduced from arXiv: 2602.17071 by Chunlei Meng, Kun Liu, Muge Qi, Rong Fu, Shuo Yin, Simon Fong, Zhaolu Kang.

**Figure 1.** Figure 1: Overview of the AdvSynGNN framework for structure-adaptive graph learning. The pipeline begins with Multi-scale Feature Synthesis, which generates node embeddings XMS by aggregating local and multi-hop contextual signals. In the core processing stage, we employ Contrastive Representation Alignment to stabilize embeddings via a self-supervised loss Lssl. Simultaneously, an Adversarial Synthesis module, cons… view at source ↗

**Figure 2.** Figure 2: Comparative embedding shifts under hybrid perturbations: (a) Original graph (b) GCN embeddings (c) [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗

**Figure 3.** Figure 3: t-SNE visualization of GAN-enhanced embeddings [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗

**Figure 4.** Figure 4: Attention patterns on heterophilous subgraph [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗

**Figure 5.** Figure 5: Visualization of GAN-induced structural perturbations: original structure and embedding, three perturbation [PITH_FULL_IMAGE:figures/full_fig_p023_5.png] view at source ↗

**Figure 6.** Figure 6: Wasserstein adversarial losses (smoothed with a 5-epoch moving average) for discriminator and generator [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗

**Figure 7.** Figure 7: Per-epoch ℓ2 gradient norms of the final convolutional block for discriminator and generator. The dashed horizontal line indicates a conservative clipping threshold of 1.0 [PITH_FULL_IMAGE:figures/full_fig_p028_7.png] view at source ↗

**Figure 8.** Figure 8: Edge-flip entropy H (Eq. 64) computed over the candidate edge set across training epochs. Higher values indicate richer perturbation diversity. H Comparative Robustness Visualization We extend [PITH_FULL_IMAGE:figures/full_fig_p028_8.png] view at source ↗

read the original abstract

Graph neural networks frequently encounter significant performance degradation when confronted with structural noise or non-homophilous topologies. To address these systemic vulnerabilities, we present AdvSynGNN, a comprehensive architecture designed for resilient node-level representation learning. The proposed framework orchestrates multi-resolution structural synthesis alongside contrastive objectives to establish geometry-sensitive initializations. We develop a transformer backbone that adaptively accommodates heterophily by modulating attention mechanisms through learned topological signals. Central to our contribution is an integrated adversarial propagation engine, where a generative component identifies potential connectivity alterations while a discriminator enforces global coherence. Furthermore, label refinement is achieved through a residual correction scheme guided by per-node confidence metrics, which facilitates precise control over iterative stability. Empirical evaluations demonstrate that this synergistic approach effectively optimizes predictive accuracy across diverse graph distributions while maintaining computational efficiency. The study concludes with practical implementation protocols to ensure the robust deployment of the AdvSynGNN system in large-scale environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

AdvSynGNN stitches adversarial synthesis and residual correction onto a transformer GNN backbone for heterophilous graphs, but the experiments do not isolate whether the added pieces deliver net gains in accuracy or efficiency.

read the letter

AdvSynGNN tries to fix the usual GNN problems on graphs that lack homophily or contain structural noise. It builds a transformer that adjusts attention using learned topological signals, adds contrastive pre-training for better starting points, runs an adversarial generator-discriminator loop to propose connectivity changes, and then applies per-node residual corrections guided by scores to refine labels over iterations. The abstract claims this combination improves accuracy across graph types without blowing up compute, and the paper supplies implementation notes for scaling it up. That assembly is the main new element; none of the pieces are invented from scratch, but the specific orchestration for node classification is presented as a working system. The experiments apparently test it on several datasets and show gains over standard baselines, which is the part that could be useful to people who need something off-the-shelf for noisy real graphs. The main weakness is that the reported results do not break out the adversarial engine's extra FLOPs or memory cost against a plain transformer GNN on the same heterophilous data. Without those controls it is hard to judge whether the efficiency claim survives or whether the extra modules simply trade one set of hyperparameters for another. The per-node thresholds and attention modulation parameters are also left as free choices, so any reader would need to see how sensitive the gains are to those choices. This paper is aimed at applied researchers who already work with GNNs on recommendation, biology, or fraud graphs and want a concrete architecture to try. It is coherent enough on its own terms to deserve a serious referee, mainly so the authors can add the missing ablations and confirm the stability claims under heterophily. I would send it out for review rather than desk-reject it.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes AdvSynGNN, a graph neural network architecture for resilient node-level representation learning on graphs with structural noise or non-homophilous topologies. It integrates multi-resolution structural synthesis and contrastive objectives for geometry-sensitive initializations, a transformer backbone that modulates attention via learned topological signals to handle heterophily, an adversarial propagation engine (generator for connectivity alterations paired with a discriminator for global coherence), and a residual correction scheme for label refinement guided by per-node confidence metrics to ensure iterative stability. The central claim is that this synergistic design optimizes predictive accuracy across diverse graph distributions while preserving computational efficiency, with practical implementation protocols provided for large-scale deployment.

Significance. If the empirical claims hold with proper validation, the work could meaningfully advance GNN robustness by providing an integrated framework that simultaneously addresses structural synthesis, heterophily adaptation, and label stability through adversarial and residual mechanisms. This would be valuable for applications on noisy or heterophilous graphs where standard GNNs degrade, potentially improving reliability in domains like social networks or molecular graphs, especially if efficiency gains are demonstrated without hidden tuning costs.

major comments (3)

[Abstract] Abstract: The central empirical claim that the synergistic approach 'effectively optimizes predictive accuracy across diverse graph distributions while maintaining computational efficiency' is stated without any reported metrics, baselines, ablation studies, error bars, or statistical tests. This absence is load-bearing because the reader's strongest claim and the paper's contribution rest entirely on these unshown evaluations.
[Abstract] Abstract: No analysis or isolation is provided for the adversarial propagation engine's computational overhead (FLOPs/memory) versus a plain transformer GNN backbone, nor any stability checks under heterophily (e.g., attention oscillation or divergent refinement). This directly impacts the efficiency claim and the weakest assumption about stable integration without extensive tuning.
[Abstract] Abstract: The per-node confidence metrics and residual correction scheme are described at a high level with no equations, convergence guarantees, or pseudocode, leaving open the risk of introduced instabilities that could undermine the claimed robustness.

minor comments (1)

[Abstract] The abstract introduces multiple novel components (e.g., 'adversarial propagation engine', 'self-corrective propagation') without referencing related prior work on adversarial GNNs or residual label propagation, which would aid context.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We are grateful to the referee for their constructive comments, which have helped us identify areas for improvement in our manuscript. We provide point-by-point responses to the major comments below.

read point-by-point responses

Referee: [Abstract] Abstract: The central empirical claim that the synergistic approach 'effectively optimizes predictive accuracy across diverse graph distributions while maintaining computational efficiency' is stated without any reported metrics, baselines, ablation studies, error bars, or statistical tests. This absence is load-bearing because the reader's strongest claim and the paper's contribution rest entirely on these unshown evaluations.

Authors: We thank the referee for this comment. The abstract is a concise summary, but we recognize the need for supporting evidence. We will revise the abstract to include specific quantitative results, such as accuracy metrics on benchmark datasets with comparisons to baselines and error bars, drawn from our experimental evaluations. revision: yes
Referee: [Abstract] Abstract: No analysis or isolation is provided for the adversarial propagation engine's computational overhead (FLOPs/memory) versus a plain transformer GNN backbone, nor any stability checks under heterophily (e.g., attention oscillation or divergent refinement). This directly impacts the efficiency claim and the weakest assumption about stable integration without extensive tuning.

Authors: We acknowledge the importance of quantifying the overhead of the adversarial propagation engine. We will add a dedicated analysis in the experimental section of the revised manuscript, including FLOPs and memory comparisons to a plain transformer GNN, as well as stability evaluations under varying heterophily levels to demonstrate stable integration. revision: yes
Referee: [Abstract] Abstract: The per-node confidence metrics and residual correction scheme are described at a high level with no equations, convergence guarantees, or pseudocode, leaving open the risk of introduced instabilities that could undermine the claimed robustness.

Authors: We appreciate this observation regarding the lack of formal details. In the revision, we will include the mathematical definitions of the per-node confidence metrics and the residual correction scheme, along with pseudocode for the iterative refinement process and a discussion of its convergence properties to support the robustness claims. revision: yes

Circularity Check

0 steps flagged

No circularity detected; architecture claims rest on empirical evaluation without self-referential derivations or fitted predictions

full rationale

The provided abstract and context describe AdvSynGNN as an integrated framework using adversarial propagation, transformer attention modulation via learned signals, and residual label correction. No equations, fitting procedures, or derivation steps are visible that would reduce any claimed prediction or result to its own inputs by construction. No self-citation chains are invoked to justify uniqueness or ansatzes. The central claims are presented as empirical outcomes across graph distributions, which are independent of any internal circular reduction. This is the expected honest non-finding for a high-level architectural description lacking visible mathematical steps.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

Only the abstract is available, so the ledger records components and assumptions implied by the prose rather than explicit derivations; several unstated premises about the stability of adversarial graph synthesis and the reliability of confidence-guided correction are required for the central claim.

free parameters (2)

per-node confidence thresholds
Used to control iterative label refinement; value and selection procedure not specified.
attention modulation parameters
Learned topological signals that adapt the transformer; fitting details absent.

axioms (2)

domain assumption Multi-resolution structural synthesis produces geometry-sensitive initializations that improve downstream learning
Invoked in the first stage of the framework description.
domain assumption Adversarial generation of connectivity alterations can be balanced by a discriminator to enforce global coherence
Central premise of the adversarial propagation engine.

invented entities (1)

adversarial propagation engine no independent evidence
purpose: Generates potential connectivity changes while a discriminator maintains coherence
New named component introduced to address structural noise.

pith-pipeline@v0.9.0 · 5475 in / 1587 out tokens · 41792 ms · 2026-05-15T21:28:33.464299+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean J_uniquely_calibrated_via_higher_derivative unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

adversarial propagation engine... residual correction scheme guided by per-node confidence metrics... heterophily-adaptive attention... spectral clipping and confidence capping
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem B.1 (Residual convergence)... contraction factor κ = max ci · ||eA||2 < 1

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

90 extracted references · 90 canonical work pages · 4 internal anchors

[1]

Expander graph propagation

Andreea Deac, Marc Lackenby, and Petar Veli ˇckovi´c. Expander graph propagation. InLearning on Graphs Conference, pages 38–1. PMLR, 2022

work page 2022
[2]

p-laplacian based graph neural networks

Guoji Fu, Peilin Zhao, and Yatao Bian. p-laplacian based graph neural networks. InInternational conference on machine learning, pages 6878–6917. PMLR, 2022

work page 2022
[3]

Nafs: a simple yet tough-to-beat baseline for graph representation learning

Wentao Zhang, Zeang Sheng, Mingyu Yang, Yang Li, Yu Shen, Zhi Yang, and Bin Cui. Nafs: a simple yet tough-to-beat baseline for graph representation learning. InInternational Conference on Machine Learning, pages 26467–26483. PMLR, 2022

work page 2022
[4]

Combining label propagation and simple models out-performs graph neural networks.arXiv preprint arXiv:2010.13993, 2020

Qian Huang, Horace He, Abhay Singh, Ser-Nam Lim, and Austin R Benson. Combining label propagation and simple models out-performs graph neural networks.arXiv preprint arXiv:2010.13993, 2020

work page arXiv 2010
[5]

Scalable and adaptive graph neural networks with self-label-enhanced training.Pattern Recognition, 160:111210, 2025

Chuxiong Sun, Jie Hu, Hongming Gu, Jinpeng Chen, Wei Liang, and Mingchuan Yang. Scalable and adaptive graph neural networks with self-label-enhanced training.Pattern Recognition, 160:111210, 2025

work page 2025
[6]

Xgcn: a library for large-scale graph neural network recommendations.Frontiers of Computer Science, 18(3):183343, 2024

Xiran Song, Hong Huang, Jianxun Lian, and Hai Jin. Xgcn: a library for large-scale graph neural network recommendations.Frontiers of Computer Science, 18(3):183343, 2024

work page 2024
[7]

What is missing for graph homophily? disentangling graph homophily for graph neural networks.Advances in Neural Information Processing Systems, 37:68406–68452, 2024

Yilun Zheng, Sitao Luan, and Lihui Chen. What is missing for graph homophily? disentangling graph homophily for graph neural networks.Advances in Neural Information Processing Systems, 37:68406–68452, 2024

work page 2024
[8]

Can large language models improve the adversarial robustness of graph neural networks? InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

Zhongjian Zhang, Xiao Wang, Huichi Zhou, Yue Yu, Mengmei Zhang, Cheng Yang, and Chuan Shi. Can large language models improve the adversarial robustness of graph neural networks? InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, pages 2008–2019, 2025

work page 2008
[9]

Mdgnn: Multi-relational dynamic graph neural network for comprehensive and dynamic stock investment prediction

Hao Qian, Hongting Zhou, Qian Zhao, Hao Chen, Hongxiang Yao, Jingwei Wang, Ziqi Liu, Fei Yu, Zhiqiang Zhang, and Jun Zhou. Mdgnn: Multi-relational dynamic graph neural network for comprehensive and dynamic stock investment prediction. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 14642–14650, 2024

work page 2024
[10]

Convolutional neural networks on graphs with fast localized spectral filtering.Advances in neural information processing systems, 29, 2016

Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering.Advances in neural information processing systems, 29, 2016

work page 2016
[11]

Semi-Supervised Classification with Graph Convolutional Networks

TN Kipf. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016
[12]

Graph Attention Networks

Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. Graph attention networks.arXiv preprint arXiv:1710.10903, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[13]

Sign: Scalable inception graph neural networks.arXiv preprint arXiv:2004.11198, 2020

Fabrizio Frasca, Emanuele Rossi, Davide Eynard, Ben Chamberlain, Michael Bronstein, and Federico Monti. Sign: Scalable inception graph neural networks.arXiv preprint arXiv:2004.11198, 2020

work page arXiv 2004
[14]

Plain transformers can be powerful graph learners.arXiv preprint arXiv:2504.12588, 2025

Liheng Ma, Soumyasundar Pal, Yingxue Zhang, Philip HS Torr, and Mark Coates. Plain transformers can be powerful graph learners.arXiv preprint arXiv:2504.12588, 2025

work page arXiv 2025
[15]

Simplifying approach to node classification in graph neural networks.Journal of Computational Science, 62:101695, 2022

Sunil Kumar Maurya, Xin Liu, and Tsuyoshi Murata. Simplifying approach to node classification in graph neural networks.Journal of Computational Science, 62:101695, 2022

work page 2022
[16]

Bridging the gap between spatial and spectral domains: A unified framework for graph neural networks.ACM Computing Surveys, 56(5):1–42, 2023

Zhiqian Chen, Fanglan Chen, Lei Zhang, Taoran Ji, Kaiqun Fu, Liang Zhao, Feng Chen, Lingfei Wu, Charu Aggarwal, and Chang-Tien Lu. Bridging the gap between spatial and spectral domains: A unified framework for graph neural networks.ACM Computing Surveys, 56(5):1–42, 2023. 11 AdvSynGNN

work page 2023
[17]

Large scale learning on non-homophilous graphs: New benchmarks and strong simple methods.Advances in neural information processing systems, 34:20887–20902, 2021

Derek Lim, Felix Hohne, Xiuyu Li, Sijia Linda Huang, Vaishnavi Gupta, Omkar Bhalerao, and Ser Nam Lim. Large scale learning on non-homophilous graphs: New benchmarks and strong simple methods.Advances in neural information processing systems, 34:20887–20902, 2021

work page 2021
[18]

Feature transportation improves graph neural networks

Moshe Eliasof, Eldad Haber, and Eran Treister. Feature transportation improves graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 11874–11882, 2024

work page 2024
[19]

Fdphormer: beyond homophily with feature- difference position encoding.ACM Transactions on Knowledge Discovery from Data, 19(5):1–19, 2025

Dong Li, Aijia Zhang, Huan Xiong, Biqing Qi, and Junqi Gao. Fdphormer: beyond homophily with feature- difference position encoding.ACM Transactions on Knowledge Discovery from Data, 19(5):1–19, 2025

work page 2025
[20]

Hgphormer: Heterophilic graph transformer.Knowledge-Based Systems, page 114031, 2025

Jianshe Wu, Yaolin Liu, Yuqian Wang, Lingjie Zhang, and Jingyi Ding. Hgphormer: Heterophilic graph transformer.Knowledge-Based Systems, page 114031, 2025

work page 2025
[21]

Long-range meta-path search on large-scale heterogeneous graphs.Advances in Neural Information Processing Systems, 37:44240–44268, 2024

Chao Li, Zijie Guo, Kun He, et al. Long-range meta-path search on large-scale heterogeneous graphs.Advances in Neural Information Processing Systems, 37:44240–44268, 2024

work page 2024
[22]

Nonlinear correct and smooth for graph-based semi-supervised learning.ACM Transactions on Knowledge Discovery from Data, 19(3):1–32, 2025

Yuanhang Shao and Xiuwen Liu. Nonlinear correct and smooth for graph-based semi-supervised learning.ACM Transactions on Knowledge Discovery from Data, 19(3):1–32, 2025

work page 2025
[23]

Mdgcl: Graph contrastive learning framework with multiple graph diffusion methods.Neural Processing Letters, 56(4):213, 2024

Yuqiang Li, Yi Zhang, and Chun Liu. Mdgcl: Graph contrastive learning framework with multiple graph diffusion methods.Neural Processing Letters, 56(4):213, 2024

work page 2024
[24]

Adversarial contrastive graph augmentation with counterfactual regularization

Tao Long, Lei Zhang, Liang Zhang, and Laizhong Cui. Adversarial contrastive graph augmentation with counterfactual regularization. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 19086–19094, 2025

work page 2025
[25]

Self-supervised graph feature enhancement and scale attention for mechanical signal node-level representation and diagnosis.Advanced Engineering Informatics, 65:103197, 2025

Xin Zhang, Jie Liu, Xi Zhang, and Yanglong Lu. Self-supervised graph feature enhancement and scale attention for mechanical signal node-level representation and diagnosis.Advanced Engineering Informatics, 65:103197, 2025

work page 2025
[26]

Graph-based semi-supervised learning: A comprehensive review.IEEE Transactions on Neural Networks and Learning Systems, 34(11):8174–8194, 2022

Zixing Song, Xiangli Yang, Zenglin Xu, and Irwin King. Graph-based semi-supervised learning: A comprehensive review.IEEE Transactions on Neural Networks and Learning Systems, 34(11):8174–8194, 2022

work page 2022
[27]

Taxonomy of benchmarks in graph representation learning

Renming Liu, Semih Cantürk, Frederik Wenkel, Sarah McGuire, Xinyi Wang, Anna Little, Leslie O’Bray, Michael Perlmutter, Bastian Rieck, Matthew Hirn, et al. Taxonomy of benchmarks in graph representation learning. In Learning on Graphs Conference, pages 6–1. PMLR, 2022

work page 2022
[28]

Constrained adaptive projection with pretrained features for anomaly detection.arXiv preprint arXiv:2112.02597, 2021

Xingtai Gui, Di Wu, Yang Chang, and Shicai Fan. Constrained adaptive projection with pretrained features for anomaly detection.arXiv preprint arXiv:2112.02597, 2021

work page arXiv 2021
[29]

Adedgedrop: Adversarial edge dropping for robust graph neural networks.IEEE Transactions on Knowledge and Data Engineering, 2025

Zhaoliang Chen, Zhihao Wu, Ylli Sadikaj, Claudia Plant, Hong-Ning Dai, Shiping Wang, Yiu-Ming Cheung, and Wenzhong Guo. Adedgedrop: Adversarial edge dropping for robust graph neural networks.IEEE Transactions on Knowledge and Data Engineering, 2025

work page 2025
[30]

Denoising structure against adversarial attacks on graph representation learning.ACM Transactions on Intelligent Systems and Technology, 16(3):1–23, 2025

Na Chen, Ping Li, Jincheng Huang, and Kai Zhang. Denoising structure against adversarial attacks on graph representation learning.ACM Transactions on Intelligent Systems and Technology, 16(3):1–23, 2025

work page 2025
[31]

Pruning spurious subgraphs for graph out-of-distribtuion generalization.arXiv preprint arXiv:2506.05957, 2025

Tianjun Yao, Haoxuan Li, Yongqiang Chen, Tongliang Liu, Le Song, Eric Xing, and Zhiqiang Shen. Pruning spurious subgraphs for graph out-of-distribtuion generalization.arXiv preprint arXiv:2506.05957, 2025

work page arXiv 2025
[32]

Unifying invariant and variant features for graph out-of-distribution via probability of necessity and sufficiency.Neural Networks, 184:107044, 2025

Xuexin Chen, Ruichu Cai, Kaitao Zheng, Zhifan Jiang, Zhengting Huang, Zhifeng Hao, and Zijian Li. Unifying invariant and variant features for graph out-of-distribution via probability of necessity and sufficiency.Neural Networks, 184:107044, 2025

work page 2025
[33]

Diffusion on graph: Augmentation of graph structure for node classification.arXiv preprint arXiv:2503.12563, 2025

Yancheng Wang, Changyu Liu, and Yingzhen Yang. Diffusion on graph: Augmentation of graph structure for node classification.arXiv preprint arXiv:2503.12563, 2025

work page arXiv 2025
[34]

Multi-scale graph diffusion convolutional network for multi-view learning.Artificial Intelligence Review, 58(6):184, 2025

Shiping Wang, Jiacheng Li, Yuhong Chen, Zhihao Wu, Aiping Huang, and Le Zhang. Multi-scale graph diffusion convolutional network for multi-view learning.Artificial Intelligence Review, 58(6):184, 2025

work page 2025
[35]

Do transformers really perform badly for graph representation?Advances in neural information processing systems, 34:28877–28888, 2021

Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, and Tie-Yan Liu. Do transformers really perform badly for graph representation?Advances in neural information processing systems, 34:28877–28888, 2021. 12 AdvSynGNN

work page 2021
[36]

Goat: A global transformer on large-scale graphs

Kezhi Kong, Jiuhai Chen, John Kirchenbauer, Renkun Ni, C Bayan Bruss, and Tom Goldstein. Goat: A global transformer on large-scale graphs. InInternational Conference on Machine Learning, pages 17375–17390. PMLR, 2023

work page 2023
[37]

Graph positional encoding via random feature propagation

Moshe Eliasof, Fabrizio Frasca, Beatrice Bevilacqua, Eran Treister, Gal Chechik, and Haggai Maron. Graph positional encoding via random feature propagation. InInternational Conference on Machine Learning, pages 9202–9223. PMLR, 2023

work page 2023
[38]

Nlgt: Neighborhood-based and label-enhanced graph transformer framework for node classification

Xiaolong Xu, Yibo Zhou, Haolong Xiang, Xiaoyong Li, Xuyun Zhang, Lianyong Qi, and Wanchun Dou. Nlgt: Neighborhood-based and label-enhanced graph transformer framework for node classification. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 12954–12962, 2025

work page 2025
[39]

Hopgat: A multi-hop graph attention network with heterophily and degree awareness.Pattern Recognition, page 112387, 2025

Han Zhang, Huan Wang, and Mingjing Han. Hopgat: A multi-hop graph attention network with heterophily and degree awareness.Pattern Recognition, page 112387, 2025

work page 2025
[40]

Mixed Precision Training

Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen, David Garcia, Boris Gins- burg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, et al. Mixed precision training.arXiv preprint arXiv:1710.03740, 2017

work page internal anchor Pith review Pith/arXiv arXiv 2017
[41]

Training Deep Nets with Sublinear Memory Cost

Tianqi Chen, Bing Xu, Chiyuan Zhang, and Carlos Guestrin. Training deep nets with sublinear memory cost. arXiv preprint arXiv:1604.06174, 2016

work page internal anchor Pith review Pith/arXiv arXiv 2016
[42]

Towards scalable and deep graph neural networks via noise masking

Yuxuan Liang, Wentao Zhang, Zeang Sheng, Ling Yang, Quanqing Xu, Jiawei Jiang, Yunhai Tong, and Bin Cui. Towards scalable and deep graph neural networks via noise masking. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 18693–18701, 2025

work page 2025
[43]

Tt-gnn: Efficient on-chip graph neural network training via embedding reformation and hardware optimization

Zheng Qu, Dimin Niu, Shuangchen Li, Hongzhong Zheng, and Yuan Xie. Tt-gnn: Efficient on-chip graph neural network training via embedding reformation and hardware optimization. InProceedings of the 56th Annual IEEE/ACM International Symposium on Microarchitecture, pages 452–464, 2023

work page 2023
[44]

Rsc: accelerate graph neural networks training via randomized sparse computations

Zirui Liu, Chen Shengyuan, Kaixiong Zhou, Daochen Zha, Xiao Huang, and Xia Hu. Rsc: accelerate graph neural networks training via randomized sparse computations. InInternational Conference on Machine Learning, pages 21951–21968. PMLR, 2023

work page 2023
[45]

Linear-time graph neural networks for scalable recommendations

Jiahao Zhang, Rui Xue, Wenqi Fan, Xin Xu, Qing Li, Jian Pei, and Xiaorui Liu. Linear-time graph neural networks for scalable recommendations. InProceedings of the ACM Web Conference 2024, pages 3533–3544, 2024

work page 2024
[46]

Accurate and scalable graph neural networks for billion-scale graphs

Juxiang Zeng, Pinghui Wang, Lin Lan, Junzhou Zhao, Feiyang Sun, Jing Tao, Junlan Feng, Min Hu, and Xiaohong Guan. Accurate and scalable graph neural networks for billion-scale graphs. In2022 IEEE 38th International Conference on Data Engineering (ICDE), pages 110–122. IEEE, 2022

work page 2022
[47]

Scgc: Self-supervised contrastive graph clustering.Neurocomputing, 611:128629, 2025

Gayan K Kulatilleke, Marius Portmann, and Shekhar S Chandra. Scgc: Self-supervised contrastive graph clustering.Neurocomputing, 611:128629, 2025

work page 2025
[48]

Dfa-gnn: Forward learning of graph neural networks by direct feedback alignment.Advances in Neural Information Processing Systems, 37: 59289–59313, 2024

Gongpei Zhao, Tao Wang, Congyan Lang, Yi Jin, Yidong Li, and Haibin Ling. Dfa-gnn: Forward learning of graph neural networks by direct feedback alignment.Advances in Neural Information Processing Systems, 37: 59289–59313, 2024

work page 2024
[49]

Brettler Liad, Berman Eden, Bartal Alon, et al. Drugnnosis-moa: Elucidating drug mechanisms as etiological or palliative with graph neural networks employing a large language model.IEEE Journal of Biomedical and Health Informatics, 2025

work page 2025
[50]

Label information enhanced fraud detection against low homophily in graphs

Yuchen Wang, Jinghui Zhang, Zhengjie Huang, Weibin Li, Shikun Feng, Ziheng Ma, Yu Sun, Dianhai Yu, Fang Dong, Jiahui Jin, et al. Label information enhanced fraud detection against low homophily in graphs. In Proceedings of the ACM Web Conference 2023, pages 406–416, 2023

work page 2023
[51]

Braingb: a benchmark for brain network analysis with graph neural networks

Hejie Cui, Wei Dai, Yanqiao Zhu, Xuan Kan, Antonio Aodong Chen Gu, Joshua Lukemire, Liang Zhan, Lifang He, Ying Guo, and Carl Yang. Braingb: a benchmark for brain network analysis with graph neural networks. IEEE transactions on medical imaging, 42(2):493–506, 2022

work page 2022
[52]

Open graph benchmark: Datasets for machine learning on graphs.Advances in neural information processing systems, 33:22118–22133, 2020

Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs.Advances in neural information processing systems, 33:22118–22133, 2020. 13 AdvSynGNN

work page 2020
[53]

Ags-gnn: Attribute-guided sampling for graph neural networks

Siddhartha Shankar Das, SM Ferdous, Mahantesh M Halappanavar, Edoardo Serra, and Alex Pothen. Ags-gnn: Attribute-guided sampling for graph neural networks. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 538–549, 2024

work page 2024
[54]

Optimal block-wise asymmetric graph construction for graph-based semi-supervised learning.Advances in Neural Information Processing Systems, 36:71135–71149, 2023

Zixing Song, Yifei Zhang, and Irwin King. Optimal block-wise asymmetric graph construction for graph-based semi-supervised learning.Advances in Neural Information Processing Systems, 36:71135–71149, 2023

work page 2023
[55]

Toward effective digraph representation learning: A magnetic adaptive propagation based approach

Xunkai Li, Daohan Su, Zhengyu Wu, Guang Zeng, Hongchao Qin, Rong-Hua Li, and Guoren Wang. Toward effective digraph representation learning: A magnetic adaptive propagation based approach. InProceedings of the ACM on Web Conference 2025, pages 2908–2923, 2025

work page 2025
[56]

Beyond graph convolution: Multimodal recommendation with topology-aware mlps

Junjie Huang, Jiarui Qin, Yong Yu, and Weinan Zhang. Beyond graph convolution: Multimodal recommendation with topology-aware mlps. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 11808–11816, 2025

work page 2025
[57]

Data-efficient graph learning

Kaize Ding. Data-efficient graph learning. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 22663–22663, 2024

work page 2024
[58]

Ogb-lsc: A large-scale challenge for machine learning on graphs.arXiv preprint arXiv:2103.09430,

Weihua Hu, Matthias Fey, Hongyu Ren, Maho Nakata, Yuxiao Dong, and Jure Leskovec. Ogb-lsc: A large-scale challenge for machine learning on graphs.arXiv preprint arXiv:2103.09430, 2021

work page arXiv 2021
[59]

Graphgan: Graph representation learning with generative adversarial nets

Hongwei Wang, Jia Wang, Jialin Wang, Miao Zhao, Weinan Zhang, Fuzheng Zhang, Xing Xie, and Minyi Guo. Graphgan: Graph representation learning with generative adversarial nets. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

work page 2018
[60]

Recipe for a general, powerful, scalable graph transformer.Advances in Neural Information Processing Systems, 35:14501–14515, 2022

Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. Recipe for a general, powerful, scalable graph transformer.Advances in Neural Information Processing Systems, 35:14501–14515, 2022

work page 2022
[61]

Graphformers: Gnn-nested transformers for representation learning on textual graph.Advances in Neural Information Processing Systems, 34:28798–28810, 2021

Junhan Yang, Zheng Liu, Shitao Xiao, Chaozhuo Li, Defu Lian, Sanjay Agrawal, Amit Singh, Guangzhong Sun, and Xing Xie. Graphformers: Gnn-nested transformers for representation learning on textual graph.Advances in Neural Information Processing Systems, 34:28798–28810, 2021

work page 2021
[62]

The impact of the mit-bih arrhythmia database.IEEE engineering in medicine and biology magazine, 20(3):45–50, 2001

George B Moody and Roger G Mark. The impact of the mit-bih arrhythmia database.IEEE engineering in medicine and biology magazine, 20(3):45–50, 2001

work page 2001
[63]

Fast global alignment kernels

Marco Cuturi. Fast global alignment kernels. InProceedings of the 28th international conference on machine learning (ICML-11), pages 929–936, 2011

work page 2011
[64]

Experimental database for detecting and diagnosing rotor broken bar in a three-phase induction motor.IEEE DataPort, 2020

Aline Elly Treml, R Andrade Flauzino, Marcelo Suetake, NA Ravazzoli Maciejewski, and N Afonso. Experimental database for detecting and diagnosing rotor broken bar in a three-phase induction motor.IEEE DataPort, 2020

work page 2020
[65]

Time-series generative adversarial networks.Advances in neural information processing systems, 32, 2019

Jinsung Yoon, Daniel Jarrett, and Mihaela Van der Schaar. Time-series generative adversarial networks.Advances in neural information processing systems, 32, 2019

work page 2019
[66]

Conditional sig-wasserstein gans for time series generation.arXiv preprint arXiv:2006.05421, 2020

Shujian Liao, Hao Ni, Lukasz Szpruch, Magnus Wiese, Marc Sabate-Vidales, and Baoren Xiao. Conditional sig-wasserstein gans for time series generation.arXiv preprint arXiv:2006.05421, 2020

work page arXiv 2006
[67]

Generative moment matching networks

Yujia Li, Kevin Swersky, and Rich Zemel. Generative moment matching networks. InInternational conference on machine learning, pages 1718–1727. PMLR, 2015

work page 2015
[68]

Rcgan: learning a generative model for arbitrary size image generation

Renato B Arantes, George V ogiatzis, and Diego R Faria. Rcgan: learning a generative model for arbitrary size image generation. InInternational Symposium on Visual Computing, pages 80–94. Springer, 2020

work page 2020
[69]

Gat-gan: A graph-attention-based time-series generative adversarial network

Srikrishna Iyer and Teng Teck Hou. Gat-gan: A graph-attention-based time-series generative adversarial network. arXiv preprint arXiv:2306.01999, 2023

work page arXiv 2023
[70]

Gan-based deep neural networks for graph representation learning.Engineering Reports, 4(11):e12517, 2022

Ming Zhao and Yinglong Zhang. Gan-based deep neural networks for graph representation learning.Engineering Reports, 4(11):e12517, 2022

work page 2022
[71]

Gan-based self-supervised message passing graph representation learning

Yining Yang, Ke Xu, and Ying Tang. Gan-based self-supervised message passing graph representation learning. Expert Systems with Applications, 251:124012, 2024. 14 AdvSynGNN

work page 2024
[72]

Att-gan: A deep learning model for dynamic network weighted link prediction

Shun Tang and Xiaoqiang Xiao. Att-gan: A deep learning model for dynamic network weighted link prediction. In2021 IEEE 3rd International Conference on Frontiers Technology of Information and Computer (ICFTIC), pages 15–20. IEEE, 2021

work page 2021
[73]

Tengan: Pure transformer encoders make an efficient discrete gan for de novo molecular generation

Chen Li and Yoshihiro Yamanishi. Tengan: Pure transformer encoders make an efficient discrete gan for de novo molecular generation. InInternational Conference on Artificial Intelligence and Statistics, pages 361–369. PMLR, 2024

work page 2024
[74]

Graph transformer gans for graph-constrained house generation

Hao Tang, Zhenyu Zhang, Humphrey Shi, Bo Li, Ling Shao, Nicu Sebe, Radu Timofte, and Luc Van Gool. Graph transformer gans for graph-constrained house generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2173–2182, 2023

work page 2023
[75]

Graph transformers for large graphs.arXiv preprint arXiv:2312.11109, 2023

Vijay Prakash Dwivedi, Yozen Liu, Anh Tuan Luu, Xavier Bresson, Neil Shah, and Tong Zhao. Graph transformers for large graphs.arXiv preprint arXiv:2312.11109, 2023

work page arXiv 2023
[76]

Sgformer: Simplifying and empowering transformers for large-graph representations.Advances in Neural Information Processing Systems, 36:64753–64773, 2023

Qitian Wu, Wentao Zhao, Chenxiao Yang, Hengrui Zhang, Fan Nie, Haitian Jiang, Yatao Bian, and Junchi Yan. Sgformer: Simplifying and empowering transformers for large-graph representations.Advances in Neural Information Processing Systems, 36:64753–64773, 2023

work page 2023
[77]

Wiki-cs: A wikipedia-based benchmark for graph neural networks.arXiv preprint arXiv:2007.02901, 2020

Péter Mernyei and C˘at˘alina Cangea. Wiki-cs: A wikipedia-based benchmark for graph neural networks.arXiv preprint arXiv:2007.02901, 2020

work page arXiv 2007
[78]

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

Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 257–266, 2019

work page 2019
[79]

Gfformer: A graph transformer for extracting all frequency information from large-scale graphs.ACM Transactions on Knowledge Discovery from Data, 2024

Qi Zhang, Mengmeng Si, Yanfeng Sun, Shaofan Wang, Junbin Gao, and Baocai Yin. Gfformer: A graph transformer for extracting all frequency information from large-scale graphs.ACM Transactions on Knowledge Discovery from Data, 2024

work page 2024
[80]

Vcr-graphormer: A mini-batch graph transformer via virtual connections.arXiv preprint arXiv:2403.16030, 2024

Dongqi Fu, Zhigang Hua, Yan Xie, Jin Fang, Si Zhang, Kaan Sancak, Hao Wu, Andrey Malevich, Jingrui He, and Bo Long. Vcr-graphormer: A mini-batch graph transformer via virtual connections.arXiv preprint arXiv:2403.16030, 2024

work page arXiv 2024

Showing first 80 references.