betaFeedback

WorksComputer Science

RGDM: Robust Graph Diffusion Model

Abstract

Diffusion models represent a powerful class of generative models known for their solid theoretical foundations and remarkable performance across diverse tasks and domains. Existing node level diffusion methods synthesize nodes and edges within a graph, but are not designed to learn a robust generator from adversarially corrupted graphs. In this study, we introduce the Robust Graph Diffusion Model (RGDM), designed to generate labeled synthetic graph structures consisting of nodes and edges that integrate seamlessly into a given graph. The RGDM consists of a Robust Graph Autoencoder (RGAE) and a Latent Diffusion Model (LDM). Leveraging an edge selection mechanism and an innovative low-rank regularization on the latent feature, the RGDM produces clean and high-quality synthetic graph structures, even when trained on graphs subject to adversarial attacks. Comprehensive experimental evaluations reveal that Graph Neural Networks (GNNs) trained on the augmented graph, which is formed by merging the original attacked graph with the synthetic graph structures, exhibit significantly improved robustness against various graph adversarial attacks in the context of semi-supervised node classification. The code of the RGDM is available at https://github.com/Statistical-Deep-Learning/RGDM.

Introduction

Building on the success of diffusion and latent diffusion models, recent works use diffusion-generated data to augment training sets [1, 2]. However, extensions to graphs [3, 4, 5, 6, 7] mainly target graph-level augmentation. Most node-level methods expand labels through mixup [8, 9, 10, 11, 12, 13] or GAN-based synthesis for imbalance [14, 15, 16, 17, 18]. Node level diffusion methods DoG [19] and LR-FGDM [20] share the autoencoder and diffusion template, but do not train the generator specifically for adversarial robustness. Moreover, GANs can be unstable and produce distributions misaligned with real data [21]. As a result, the challenge of generating synthetic graph structures (SGS) within a given graph, which are the synthetic nodes and their corresponding edges, in a robust manner to enhance node classification under attacks, remains underexplored. In this work, we tackle the problem of node-level graph generation and propose the Robust Graph Diffusion Model (RGDM), designed to synthesize high-quality graph structures within a single graph, even under adversarial attacks. These synthetic structures include both synthetic nodes and edges connecting them within the same graph, forming an augmented graph as illustrated in Figure 1. Robust GNNs will be trained on the augmented graph, and superior performance is expected due to the enlarged set of labeled nodes.

Figure 1. Illustration of the 2-hop neighborhood of a real node and a synthetic node in the augmented graph. Figure 5 in Section F.10 illustrates the t-SNE visualization of the augmented graph.

Generating high-quality SGS in adversarial scenarios for improved performance is challenging, important, and unaddressed in the robust graph learning literature. To address this challenge, RGDM integrates a Robust Graph Autoencoder (RGAE) with a Latent Diffusion Model (LDM). The RGAE incorporates an edge selection mechanism and a low-rank regularization on the latent features to counteract adversarial perturbations in the attacked graph. The low-rank regularization is motivated by the Low Frequency Property (LFP) [22, 23, 24, 25, 26, 19]: in the evaluated latent spaces, label information is more concentrated in leading eigendirections than adversarial perturbations (Section F.1). Theorem 3.1 provides a complementary, conditional justification through the spectral complexity of a bounded linear probe on this latent space. It is not a guarantee for the downstream GNN trained on the augmented graph.

Figure 1 illustrates the 2-hop neighborhood of a real node and a synthetic node that have similar node features, evidenced by the t-SNE visualization in Figure 5 in Section F.10 of the appendix. The fidelity evaluations in Table 13 and Table 14 compare synthetic structures with clean reference graphs through Frechet Node Distance (FND) and Frechet Edge Distance (FED). RGDM has lower distances than vanilla DDPM [27] and the compared node augmentation methods. On Pubmed under 25% Metattack, its FND/FED are 3.95/5.34, compared with 3.79/5.15 for the clean reference. These are empirical fidelity measurements. Clean graphs are used for evaluation, not for training RGDM. The augmentation pipeline retains the original attacked adjacency. Edge selection acts inside the generator, reducing the reproduction of unreliable connections. The generated labeled structures supply additional supervision and relatively clean message passing paths. On Citeseer, adding 785 synthetic edges changes the normalized perturbation density from 25% to approximately 20.6%. This is a denominator effect, not removal of the original perturbations or a causal robustness guarantee (Section F.4).

DataCoraCiteseerPolblogsPubmed
Original Attacked Graph12.10±0.457.53±0.228.26±0.376.11±0.38
iGraphMix [12]13.44±0.339.02±0.4210.75±0.447.65±0.38
ImGAGN [42]13.10±0.387.96±0.429.86±0.436.95±0.27
Graphsmote [43]13.52±0.428.95±0.2210.60±0.277.45±0.51
SNS [45]13.35±0.427.88±0.279.30±0.856.74±0.41
Baseline DDPM13.28±0.408.82±0.2610.25±0.397.25±0.41
RGDM (w/o Robustness Loss)11.17±0.466.15±0.318.06±0.445.89±0.35
RGDM (w/o Node Robustness Loss)10.02±0.385.89±0.286.95±0.355.01±0.36
RGDM (w/o Edge Robustness Loss)9.67±0.275.18±0.496.65±0.255.11±0.38
RGDM8.16±0.294.44±0.324.35±0.183.95±0.43
Original Clean Graph7.90±0.314.28±0.254.16±0.273.79±0.35

Table 13. Frechet Node Distance (FND) to the nodes in the base partition of the original clean graph. The mean and standard deviation of the FND scores computed with 10 different random splits of the base partition and the test partition in the original clean graph are reported. The evaluation is performed for Metattack with a perturbation rate of 25%. The FND score for the original clean graph is computed between the nodes in the test partition and the nodes in the base partition of the original clean graph.

Table 14: Frechet Edge Distance (FED) to the edges in the base partition of the original clean graph. The mean and standard deviation of the FED scores computed with 10 different random splits of the base partition and the test partition in the original clean graph are reported. The evaluation is performed for Metattack with a perturbation rate of 25%. The FED score for the original clean graph is computed between the edges in the test partition and the edges in the base partition of the original clean graph.

DataCoraCiteseerPolblogsPubmed
Original Attacked Graph10.32±0.427.83±0.288.92±0.396.53±0.37
iGraphMix [12]11.25±0.288.42±0.439.63±0.447.11±0.27
ImGAGN [42]11.05±0.388.04±0.268.21±0.517.16±0.79
Graphsmote [43]11.13±0.298.16±0.519.32±0.417.18±0.44
SNS [45]11.05±0.388.04±0.268.21±0.517.16±0.79
Baseline DDPM (K=⌊dave/4⌋K = \lfloor d_{\mathrm{ave}}/4 \rfloor)12.03±0.458.72±0.3110.77±0.407.93±0.42
Baseline DDPM (K=⌊dave/2⌋K = \lfloor d_{\mathrm{ave}}/2 \rfloor)11.27±0.388.49±0.279.32±0.357.28±0.34
Baseline DDPM (K=⌊dave⌋K = \lfloor d_{\mathrm{ave}} \rfloor)10.39±0.407.96±0.308.97±0.376.83±0.33
Baseline DDPM (K=⌈2×dave⌉K = \lceil2 \times d_{\mathrm{ave}} \rceil)10.47±0.367.93±0.299.21±0.366.91±0.35
Baseline DDPM (K=⌈4×dave⌉K = \lceil4 \times d_{\mathrm{ave}} \rceil)10.88±0.418.01±0.329.38±0.387.08±0.38
RGDM (w/o Robustness Loss)10.07±0.397.64±0.268.74±0.336.27±0.31
RGDM (w/o Node Robustness Loss)8.87±0.356.73±0.257.93±0.315.78±0.32
RGDM (w/o Edge Robustness Loss)8.98±0.346.82±0.277.72±0.295.91±0.30
RGDM8.34±0.295.28±0.226.61±0.255.34±0.28
Original Clean Graph8.03±0.285.10±0.236.31±0.245.15±0.25

Table 14. Frechet Edge Distance (FED) to the edges in the base partition of the original clean graph. The mean and standard deviation of the FED scores computed with 10 different random splits of the base partition and the test partition in the original clean graph are reported. The evaluation is performed for Metattack with a perturbation rate of 25%. The FED score for the original clean graph is computed between the edges in the test partition and the edges in the base partition of the original clean graph.

Figure 5. Illustration of the augmented graph after adding the synthetic graph structures into the original graph on Cora. The RGDM is trained on Cora under Metattack with a perturbation rate of 25%. The figure on the right illustrates the t-SNE visualization of the node features for both the real nodes in the original clean graph and the synthetic nodes. The figure on the left illustrates the 2-hop neighborhood of a real node and a synthetic node that have similar node features, evidenced by the t-SNE visualization in the augmented graph. It is observed that the synthetic node shares the same neighbors as the real node.

Difference from diffusion based purification. Diffusion models have also been used for data and graph purification [28, 29, 30, 31, 32]. Purifying an observed graph and generating additional labeled nodes with incident edges are different tasks. Graph level purification methods do not directly provide the node level augmentation interface of RGDM. Conversely, vanilla node generation trained on a corrupted graph can reproduce its noise. RGDM trains its robust generator without paired clean graphs. Section F.13 compares GDDM and DiffSP preprocessing with RGDM augmentation using the same SG-GSR backbone, while Section F.12 separately evaluates a purification adaptation of RGDM.

Contributions. First, RGDM learns a robust node level generator through edge selection, sparse hierarchical reconstruction, and low rank regularization of the latent features used by diffusion. Its Sparse Hierarchical Edge Decoder (SHED) evaluates selected clusters instead of globally scoring every candidate neighbor. Controlled comparisons with DoG and LR-FGDM, componentwise ablations, and wall clock measurements distinguish these contributions from the shared autoencoder and diffusion architecture. Second, RGDM augments existing classifiers without changing their architecture or objective. We evaluate its incremental benefit with each backbone fixed, including SG-GSR [33], and report paired variability. The gains over this strong backbone are modest in absolute magnitude, but consistent in the tested settings. Fidelity, heterophily, and adaptive attack evaluations support this empirical conclusion. The linear probe theorem motivates the latent regularizer and does not establish an end to end robustness guarantee.

Data Augmentation for Graph Learning. To improve the performance of GNNs, node-level graph data augmentation methods have been studied to modify the structure [34, 35, 36, 37, 38], features [39, 40, 41], or node labels [39, 40, 41]. Studies have demonstrated that increasing the number of labeled nodes based on Mix-up strategies [8, 9, 10, 11, 12, 13] can greatly enhance the performance of the GNNs. Recent works address node and edge imbalance by generating synthetic nodes and edges [42, 43, 44, 45, 46], but predominantly rely on Generative Adversarial Networks (GANs). However, the computer vision literature [21] has demonstrated that GANs often exhibit instability during training and that the distribution of synthetic data generated by GANs poorly matches the distribution of real data. Graph level diffusion models [4, 3, 5, 6, 7] generate entire graphs. DoG [19] and LR-FGDM [20] instead generate nodes and associated edges within an existing graph, as does RGDM. DoG reconstructs observed neighborhoods with a Bi-Level Neighborhood Decoder (BLND) and applies low rank regularization to its downstream GNN. LR-FGDM addresses few classification using a Hierarchical Graph Autoencoder (HGAE) with prototypical regularization, followed by a low rank transductive classifier. RGDM places edge selection, a sparsification objective, and low rank regularization inside the generator. Its TNN penalty acts on the latent features used by diffusion. SHED reconstructs selected sparse neighborhoods rather than all observed connections. The resulting augmented graph can be used with GCN, Pro-GNN, or SG-GSR. Table 4 and Table 5 compare these methods with a fixed backbone. Table 33 additionally compares hierarchical reconstruction variants.

MethodCoraCiteseerPolblogsPubmed
GCN47.5356.9449.2375.50
iGraphMix (GCN) [12]53.2557.1151.6576.10
ImGAGN (GCN) [42]53.5856.8950.9575.32
Graphsmote (GCN) [43]55.1058.0652.4476.25
SNS (GCN) [45]58.3559.1754.0378.28
DoG (GCN) [19]59.4060.1054.7079.20
LR-FGDM (GCN) [20]58.9059.6054.2078.90
RGDM (GCN)61.4262.8456.4781.52

Table 4. Comparisons between RGDM and existing node-level graph augmentation methods. The study is performed for Metattack with a perturbation rate of 25%.

ConfigurationCoraCiteseerPolblogsPubmed
SG-GSR (no augmentation)77.872.787.887.2
DoG augmentation (BLND)75.671.085.385.1
+ Edge selection76.672.186.385.9
+ Latent TNN (RGDM with BLND)77.473.187.286.6
+ SHED (full RGDM)79.175.388.888.6

Table 5. Successive generator components. SG-GSR and 25% Metattack are fixed. Accuracy in percent.

DatasetCoraCiteseerPolblogsPubmed
RGDM (HGAE from LR-FGDM)76.872.586.786.0
RGDM (BLND)77.473.187.286.6
RGDM79.175.388.888.6

Table 33. Hierarchical reconstruction comparison: SHED, BLND from DoG [19], and the HGAE variant from LR-FGDM [20]. SG-GSR is used as the base GNN under Metattack with 25% perturbation.

Graph Adversarial Attacks and Defense. Despite the success of GNNs in graph-structured tasks [47, 48], recent studies reveal their vulnerability to adversarial attacks [49]. Depending on the attack objective, threat models are categorized as: targeted attacks [50], which mislead predictions on specific nodes, and untargeted attacks [51, 52], which reduce overall model accuracy. To defend against these threats, robust learning strategies are grouped into three categories: adversarial training, graph processing, and model robustification. Adversarial training [53, 54] augments training data with crafted adversarial samples. Graph processing [55, 56, 57, 58] seeks to denoise perturbed graphs. Model robustification [59, 60, 61, 62, 63, 64, 65, 33] enhances model resilience.

Methods

Preliminaries of the Attributed Graph

An attributed graph of NN nodes is represented by G=(V,X,A)\mathcal{G} = (\mathcal{V}, \mathbf{X}, \mathbf{A}), where V={v1,v2,…,vN}\mathcal{V} = \{v_1, v_2, \ldots, v_N\} and E⊆V×V\mathcal{E} \subseteq\mathcal{V} \times\mathcal{V} denote the nodes and edges respectively. X∈RN×D\mathbf{X} \in\mathbb{R}^{N \times D} contains the node attributes. Xi∈RD\mathbf{X}_i \in\mathbb{R}^{D}, the ii-th row of X\mathbf{X}, denotes the attributes of the ii-th node. A∈{0,1}N×N\mathbf{A} \in\{0,1\}^{N \times N} is the adjacency matrix of the graph G\mathcal{G}. A~=A+I\widetilde{\mathbf{A}} = \mathbf{A} + \mathbf{I} is the adjacency matrix of a graph with self-loops added. D~\widetilde{\mathbf{D}} denotes the diagonal degree matrix of A~\widetilde{\mathbf{A}}. We use N(i)={j∣A~i,j=1}\mathcal{N}(i) = \{j \mid\widetilde{A}_{i,j} = 1\} to denote the set of nodes connected to viv_i. Z∈RN×D′\mathbf{Z} \in\mathbb{R}^{N \times D'} contains the latent features of all the nodes in the graph, where D′D' is the latent dimension. Let VL\mathcal{V}_{L} and VU\mathcal{V}_{U} denote the set of labeled nodes and unlabeled test nodes, respectively, and ∣VL∣=m|\mathcal{V}_{L}| = m, ∣VU∣=u|\mathcal{V}_{U}| = u. Let YL={yi∣vi∈VL,yi∈[C]}Y_{L} = \{y_i \mid v_i \in\mathcal{V}_{L}, y_i \in[C]\} denote the labels of nodes in VL\mathcal{V}_{L}, where CC is the number of classes. Let Y=[y1;y2;…yN]∈RN×C\mathbf{Y} = [y_1; y_2; \ldots y_N] \in\mathbb{R}^{N \times C} be the ground truth label matrix of all the nodes in G\mathcal{G}, where yiy_i is the one-hot label vector of node viv_i. Let u∈RN\mathbf{u} \in\mathbb{R}^{N} be a vector. We use [u]A[\mathbf{u}]_{A} to denote a vector formed by elements of u\mathbf{u} with indices in AA for A⊆[N]A \subseteq[N]. If u\mathbf{u} is a matrix, then [u]A[\mathbf{u}]_{A} denotes a submatrix formed by rows of u\mathbf{u} with row indices in AA. ∥⋅∥F\|\cdot\|_{F} denotes the Frobenius norm of a matrix, and ∥⋅∥p\|\cdot\|_{p} denotes the pp-norm of a vector.

Robust Graph Autoencoder (RGAE)

In this section, we propose the Robust Graph Autoencoder (RGAE) capable of encoding node attributes and associated edges into continuous latent features and decoding these features to reconstruct the node attributes and edges in the graph. Beyond reconstruction, the RGAE is designed to be robust to adversarial perturbations in the node attributes and the edges in the graph, enhancing the resilience against potential graph adversarial attacks. We encode each node’s attributes together

with its neighborhood. Positional embeddings give Xj′=Xj+pos⁡(j)X'_j = X_j + \operatorname{pos}(j), and GAT layers [66] aggregate these features. For j∈N(i)j \in\mathcal{N}(i), the attention weight is Bij=exp⁡(aij)∑ℓ∈N(i)exp⁡(aiℓ)B_{ij} = \frac{\exp(a_{ij})}{\sum_{\ell\in\mathcal{N}(i)} \exp(a_{i\ell})}, where aij=σ((Xi′Qk∥Xj′Q)V)a_{ij} = \sigma((X'_iQ^k \mathbin{\|} X'_jQ)V), ∥\mathbin{\|} denotes concatenation, σ\sigma is the sigmoid function, and QQ, VV are learnable matrices. Attention is restricted to observed neighbors. To suppress unreliable edges during RGAE training, a learned binary mask SS selects connections. We use the sigmoid relaxation S^ij=σ(θij/τ)\widehat{S}_{ij} = \sigma(\theta_{ij}/\tau) associated with binary Gumbel Softmax training [67, 68]. A linear layer applied to Xi′∥Xj′X'_i \mathbin{\|} X'_j predicts θij\theta_{ij}. The forward pass uses Sij=1{S^ij>1/2}S_{ij} = \mathbb{1}\{\widehat{S}_{ij} > 1/2\}, while a straight through estimator differentiates through S^ij\widehat{S}_{ij}. We use τ=0.5\tau= 0.5 unless stated otherwise. The selected neighbor aggregation is Zi′=ReLU⁡(∑j∈N(i)SijBijXj′W)Z'_i = \operatorname{ReLU}\left(\sum_{j \in\mathcal{N}(i)} S_{ij}B_{ij}X'_jW\right), with learnable WW. Two GAT layers yield the neighborhood representation. Concatenating it with an MLP of the node attributes gives Zi=Zi′∥f(Xi)Z_i = Z'_i \mathbin{\|} f(X_i). The mask changes the generator’s training representation, not the original adjacency retained for downstream augmentation.

References

  1. [1]Brandon Trabucco, Kyle Doherty, Max A Gurinas, and Ruslan Salakhutdinov. Effective data augmentation with diffusion models. In The Twelfth International Conference on Learning Representations, 2024.arxiv.org/abs/2302.07944
  2. [2]Shekoofeh Azizi, Simon Kornblith, Chitwan Saharia, Mohammad Norouzi, and David J Fleet. Synthetic data from diffusion models improves imagenet classification. Transactions on Machine Learning Research, 2023.arxiv.org/abs/2304.08466
  3. [3]Chenhao Niu, Yang Song, Jiaming Song, Shengjia Zhao, Aditya Grover, and Stefano Ermon. Permutation invariant graph generation via score-based generative modeling. In The 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020, 26-28 August 2020, Online [Palermo, Sicily, Italy], volume 108 of Proceedings of Machine Learning Research, pages 4474–4484. PMLR, 2020.arxiv.org/abs/2003.00638
  4. [4]Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In 9th International Conference on Learning Representations, ICLR, 2021.arxiv.org/abs/2011.13456
  5. [5]Kilian Konstantin Haefeli, Karolis Martinkus, Nathanaël Perraudin, and Roger Wattenhofer. Diffusion models for graphs benefit from discrete state spaces. CoRR, abs/2210.01549, 2022.arxiv.org/abs/2210.01549
  6. [6]Clément Vignac, Igor Krawczuk, Antoine Siraudin, Bohan Wang, Volkan Cevher, and Pascal Frossard. DiGress: Discrete denoising diffusion for graph generation. In The Eleventh International Conference on Learning Representations, ICLR 2023, 2023.arxiv.org/abs/2209.14734
  7. [7]Stratis Limnios, Praveen Selvaraj, Mihai Cucuringu, Carsten Maple, Gesine Reinert, and Andrew Elliott. Sagess: Sampling graph denoising diffusion model for scalable graph generation. CoRR, abs/2306.16827, 2023.arxiv.org/abs/2306.16827
  8. [8]Xiaotian Han, Zhimeng Jiang, Ninghao Liu, and Xia Hu. G-mixup: Graph data augmentation for graph classification. In International Conference on Machine Learning (ICML), pages 8230–8248. PMLR, 2022.arxiv.org/abs/2202.07179
  9. [9]Yiwei Wang, Wei Wang, Yuxuan Liang, Yujun Cai, and Bryan Hooi. Mixup for node and graph classification. In Proceedings of the Web Conference 2021, pages 3663–3674, 2021.DOI
  10. [10]Vikas Verma, Meng Qu, Kenji Kawaguchi, Alex Lamb, Yoshua Bengio, Juho Kannala, and Jian Tang. Graphmix: Improved training of gnns for semi-supervised learning. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educational Advances in Artificial Intelligence, EAAI 2021, Virtual Event, February 2-9, 2021, pages 10024–10032. AAAI Press, 2021.arxiv.org/abs/1909.11715
  11. [11]Wentao Zhao, Qitian Wu, Chenxiao Yang, and Junchi Yan. Geomix: Towards geometry-aware data augmentation. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 4500–4511, 2024.DOI
  12. [12]Jongwon Jeong, Hoyeop Lee, Hyui Geon Yoon, Beomyoung Lee, Junhee Heo, Geonsoo Kim, and Kim Jin Seon. igraphmix: Input graph mixup method for node classification. In The Twelfth International Conference on Learning Representations, 2024.
  13. [13]Weigang Lu, Ziyu Guan, Wei Zhao, Yaming Yang, and Long Jin. Nodemixup: Tackling under-reaching for graph neural networks. In Michael J. Wooldridge, Jennifer G. Dy, and Sriram Natarajan, editors, Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence, IAAI 2024, Fourteenth Symposium on Educational Advances in Artificial Intelligence, EAAI 2014, February 20-27, 2024, Vancouver, Canada, pages 14175–14183. AAAI Press, 2024.arxiv.org/abs/2312.13032
  14. [14]Nan Jia, Xiaolin Tian, Wenxing Gao, and Licheng Jiao. Deep graph-convolutional generative adversarial network for semi-supervised learning on graphs. Remote Sensing, 15(12):3172, 2023.DOI
  15. [15]Cheng Wu, Chaokun Wang, Jingcao Xu, Ziyang Liu, Kai Zheng, Xiaowei Wang, Yang Song, and Kun Gai. Graph contrastive learning with generative adversarial network. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 2721–2730, 2023.DOI
  16. [16]Hongwei Wang, Jia Wang, Jialin Wang, Miao Zhao, Weinan Zhang, Fuzheng Zhang, Xing Xie, and Minyi Guo. Graphghan: Graph representation learning with generative adversarial nets. In Sheila A. McIlraith and Kilian Q. Weinberger, editors, Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), the 30th innovative Applications of Artificial Intelligence (IAAI-18), and the 8th AAAI Symposium on Educational Advances in Artificial Intelligence (EAAI-18), New Orleans, Louisiana, USA, February 2-7, 2018, pages 2508–2515. AAAI Press, 2018.arxiv.org/abs/1711.08267
  17. [17]Ming Liang, Yao Meng, Jiyu Wang, David L Lubkeman, and Ning Lu. Feedergan: Synthetic feeder generation via deep graph adversarial nets. IEEE Transactions on Smart Grid, 12(2):1163–1173, 2020.DOI
  18. [18]Carl Yang, Peiye Zhuang, Wenhan Shi, Alan Luu, and Pan Li. Conditional structure generation through graph variational generative adversarial nets. Advances in Neural Information Processing Systems (NeurIPS), 32, 2019.
  19. [19]Yancheng Wang, Changyu Liu, and Yingzhen Yang. Diffusion on graph: Augmentation of graph structure for node classification. Trans. Mach. Learn. Res., 2025, 2025.arxiv.org/abs/2503.12563
  20. [20]Yancheng Wang, Chengshuai Zhao, Dongfang Sun, Huan Liu, and Yingzhen Yang. Low-rank few-shot node classification by node-level graph diffusion. In The Fourteenth International Conference on Learning Representations, 2026.
  21. [21]Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in Neural Information Processing Systems (NeurIPS), 34:8780–8794, 2021.arxiv.org/abs/2105.05233
  22. [22]Nasim Rahaman, Aristide Baratin, Devansh Arpit, Felix Draxler, Min Lin, Fred Hamprecht, Yoshua Bengio, and Aaron Courville. On the spectral bias of neural networks. In International Conference on Machine Learning, volume 97, pages 5031–5310. PMLR, 09–15 Jun 2019.arxiv.org/abs/1806.08734
  23. [23]Sanjeev Arora, Simon S. Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. In International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 322–332. PMLR, 2019.arxiv.org/abs/1901.08584
  24. [24]Yuan Cao, Zhiying Fang, Yue Wu, Ding-Xuan Zhou, and Quanquan Gu. Towards understanding the spectral bias of deep learning. In Zhi-Hua Zhou, editor, International Joint Conference on Artificial Intelligence, pages 2205–2211. ijcai.org, 2021.arxiv.org/abs/1912.01198
  25. [25]Moulik Choraria, Leello Tadesse Dadi, Grigorios Chrysos, Julien Mairal, and Volkan Cevher. The spectral bias of polynomial neural networks. In International Conference on Learning Representations. OpenReview.net, 2022.arxiv.org/abs/2202.13473
  26. [26]Yancheng Wang, Rajeev Goel, Utkarsh Nath, Alvin C. Silva, Teresa Wu, and Yingzhen Yang. Learning low-rank feature for thorax disease classification. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024, 2024.arxiv.org/abs/2404.18933
  27. [27]Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, 2020.arxiv.org/abs/2006.11239
  28. [28]Xin He, Wenqi Fan, Yili Wang, Chengyi Liu, Rui Miao, Xin Juan, and Xin Wang. Graph defense diffusion model. In Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 1, pages 416–427, 2026.arxiv.org/abs/2501.11568
  29. [29]Jiayi Luo, Qingyun Sun, Haonan Yuan, Xingcheng Fu, and Jianxin Li. Robust graph learning against adversarial evasion attacks via prior-free diffusion-based structure purification. In Proceedings of the ACM on Web Conference 2025, WWW 2025, Sydney, NSW, Australia, 28 April 2025- 2 May 2025, pages 2098–2110. ACM, 2025.arxiv.org/abs/2502.05000
  30. [30]Chu Zhao, Enneng Yang, Yuliang Liang, Jianzhe Zhao, Guibing Guo, and Xingwei Wang. Distributionally robust graph out-of-distribution recommendation via diffusion model. In Proceedings of the ACM on Web Conference 2025, WWW 2025, Sydney, NSW, Australia, 28 April 2025- 2 May 2025, pages 2018–2031. ACM, 2025.arxiv.org/abs/2501.15555
  31. [31]Ganglin Xie, Haobo Lu, Xinze Zhang, and Kun He. Advancing SAR image robustness: Integrating diffusion models for adversarial purification and speckle noise suppression. In 2025 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2025, Hyderabad, India, April 6-11, 2025, pages 1–5. IEEE, 2025.DOI
  32. [32]Xinquan Chen, Xitong Gao, Juanjuan Zhao, Kejiang Ye, and Cheng-Zhong Xu. Addiffuser: Natural adversarial example synthesis with diffusion models. In IEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, October 1-6, 2023, pages 4539–4549. IEEE, 2023.DOI
  33. [33]Yeonjun In, Kanghoon Yoon, Kibum Kim, Kijung Shin, and Chanyoung Park. Self-guided robust graph structure refinement. In Proceedings of the ACM Web Conference 2024, pages 697–708, 2024.arxiv.org/abs/2402.11837
  34. [34]Johannes Gasteiger, Stefan Weißenberger, and Stephan Günnemann. Diffusion improves graph learning. Advances in neural information processing systems, 32, 2019.arxiv.org/abs/1911.05485
  35. [35]Tong Zhao, Yozen Liu, Leonardo Neves, Oliver J. Woodford, Meng Jiang, and Neil Shah. Data augmentation for graph neural networks. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applications of Artificial Intelligence, IAAI 2021, The Eleventh Symposium on Educational Advances in Artificial Intelligence, EAAI 2021, Virtual Event, February 2-9, 2021, pages 10115–1023. AAAI Press, 2021.arxiv.org/abs/2006.06830
  36. [36]Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph convolutional networks on node classification. In International Conference on Learning Representations (ICLR). OpenReview.net, 2020.arxiv.org/abs/1907.10903
  37. [37]Wenzheng Feng, Yuxiao Dong, Tinglin Huang, Ziqi Yin, Xu Cheng, Evgeny Kharlamov, and Jie Tang. GRAND+: scalable graph random neural networks. In WWW ’22: The ACM Web Conference 2022, Virtual Event, Lyon, France, April 25 - 29, 2022, pages 3248–3258. ACM, 2022.arxiv.org/abs/2203.06389
  38. [38]Yurui Lai, Xiaoyang Lin, Renchi Yang, and Hongtao Wang. Efficient topology-aware data augmentation for high-degree graph neural networks. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 1463–1473, 2024.arxiv.org/abs/2406.05482
  39. [39]Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. Graph contrastive learning with augmentations. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin, editors, Advances in Neural Information Processing Systems (NeurIPS), 2020.arxiv.org/abs/2010.13902
  40. [40]Kezhi Kong, Guohao Li, Mucong Ding, Zuxuan Wu, Chen Zhu, Bernard Ghanem, Gavin Taylor, and Tom Goldstein. Robust optimization as data augmentation for large-scale graphs. In The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 60–69. IEEE, 2022.arxiv.org/abs/2010.09891
  41. [41]Amitoz Azad and Yuan Fang. A learned generalized geodesic distance function-based approach for node feature augmentation on graphs. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 49–58, 2024.DOI
  42. [42]Liang Ou, Huaisheng Zhu, Ruiqi Zheng, Yuhui Shi, and Hongzhi Yin. Imagnn: Imbalanced network embedding via generative adversarial graph networks. In Feida Zhu, Beng Chin Ooi, and Chunyan Miao, editors, KDD ’21: The 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, Singapore, August 14-18, 2021, pages 1390–1398. ACM, 2021.arxiv.org/abs/2106.02817
  43. [43]Tianxiang Zhao, Xiang Zhang, and Suhang Wang. Graphsmote: Imbalanced node classification on graphs with graph neural networks. In Proceedings of the 14th ACM international conference on web search and data mining, pages 833–841, 2021.DOI
  44. [44]Hung Chun Hsu, Ting-Le Li, Bo-Jun Wu, Ming-Yi Hong, Che Lin, and Chih-Yu Wang. Fincgan: A gan framework of imbalanced node classification on heterogeneous graph neural network. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 5750–5754. IEEE, 2024.
  45. [45]Xinyi Gao, Wentao Zhang, Tong Chen, Junliang Yu, Hung Quoc Viet Nguyen, and Hongzhi Yin. Semantic-aware node synthesis for imbalanced heterogeneous information networks. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pages 545–555, 2023.arxiv.org/abs/2302.14061
  46. [46]Hung Chun Hsu, Bo-Jun Wu, Ming-Yi Hong, Che Lin, and Chih-Yu Wang. A GAN approach for node embedding in heterogeneous graphs using subgraph sampling. CoRR, abs/2312.06519, 2023.arxiv.org/abs/2312.06519
  47. [47]Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings. OpenReview.net, 2017.arxiv.org/abs/1609.02907
  48. [48]Muhan Zhang and Yixin Chen. Link prediction based on graph neural networks. In Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018, 2018.arxiv.org/abs/1802.09691
  49. [49]Enyan Dai, Tianxiang Zhao, Huaisheng Zhu, Junjie Xu, Zhimeng Guo, Hui Liu, Jiliang Tang, and Suhang Wang. A comprehensive survey on trustworthy graph neural networks: Privacy, robustness, fairness, and explainability. CoRR, abs/2204.08570, 2022.DOI
  50. [50]Daniel Zügner, Amir Akbarnejad, and Stephan Günnemann. Adversarial attacks on neural networks for graph data. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pages 2847–2856, 2018.DOI
  51. [51]Daniel Zügner and Stephan Günnemann. Adversarial attacks on graph neural networks via meta learning. In International Conference on Learning Representations (ICLR), 2019.arxiv.org/abs/1902.08412
  52. [52]Yiwei Sun, Suhang Wang, Xianfeng Tang, Tsung-Yu Hsieh, and Vasant Honavar. Adversarial attacks on graph neural networks via node injections: A hierarchical reinforcement learning approach. In Proceedings of the Web Conference 2020, pages 673–683, 2020.DOI
  53. [53]Jintang Li, Jiaying Peng, Liang Chen, Zibin Zheng, Tingting Liang, and Qing Ling. Spectral adversarial training for robust graph neural network. IEEE Transactions on Knowledge and Data Engineering, 2022.arxiv.org/abs/2211.10896
  54. [54]Fuli Feng, Xiangnan He, Jie Tang, and Tat-Seng Chua. Graph adversarial training: Dynamically regularizing based on graph structure. IEEE Transactions on Knowledge and Data Engineering, 2019.arxiv.org/abs/1902.08226
  55. [55]Huijun Wu, Chen Wang, Yuriy Tyshetskiy, Andrew Docherty, Kai Lu, and Liming Zhu. Adversarial examples for graph data: Deep insights into attack and defense. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI, 2019.arxiv.org/abs/1903.01610
  56. [56]Negin Entezari, Saba A Al-Sayouri, Amirali Darvishzadeh, and Evangelos E Papalexakis. All you need is low (rank) defending against adversarial attacks on graphs. In Proceedings of the 13th International Conference on Web Search and Data Mining, pages 169–177, 2020.
  57. [57]Wei Jin, Yao Ma, Xiaorui Liu, Xianfeng Tang, Suhang Wang, and Jiliang Tang. Graph structure learning for robust graph neural networks. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, pages 66–74, 2020.arxiv.org/abs/2005.10203
  58. [58]Runlin Lei, Zhen Wang, Yaliang Li, Bolin Ding, and Zhewei Wei. Evennet: Ignoring odd-hop neighbors improves robustness of graph neural networks. Advances in Neural Information Processing Systems, 2022.arxiv.org/abs/2205.13892
  59. [59]Beini Xie, Heng Chang, Ziwei Zhang, Xin Wang, Daixin Wang, Zhiqiang Zhang, Rex Ying, and Wenwu Zhu. Adversarially robust neural architecture search for graph neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8143–8152, 2023.arxiv.org/abs/2304.04168
  60. [60]T Konstantin Rusch, Ben Chamberlain, James Rowbottom, Siddhartha Mishra, and Michael Bronstein. Graph-coupled oscillator networks. In International Conference on Machine Learning, pages 18888–18909. PMLR, 2022.arxiv.org/abs/2202.02296
  61. [61]Yang Song, Qiyu Kang, Sijie Wang, Kai Zhao, and Wee Peng Tay. On the robustness of graph neural diffusion to topology perturbations. Advances in Neural Information Processing Systems, 35:6384–6396, 2022.arxiv.org/abs/2209.07754
  62. [62]Kai Zhao, Qiyu Kang, Yang Song, Rui She, Sijie Wang, and Wee Peng Tay. Adversarial robustness in graph neural networks: A hamiltonian approach. Advances in Neural Information Processing Systems, 2023.arxiv.org/abs/2310.06396
  63. [63]Yaning Jia, Dongmian Zou, Hongfei Wang, and Hai Jin. Enhancing node-level adversarial defenses by lipschitz regularization of graph neural networks. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 951–963, 2023.DOI
  64. [64]Songtao Liu, Jinghui Chen, Tianfan Fu, Lu Lin, Marinka Zitnik, and Dinghao Wu. Graph adversarial diffusion convolution. In International Conference on Machine Learning, 2024.arxiv.org/abs/2406.02059
  65. [65]Yassine Abbahaddou, Sofiane Ennadir, Johannes F. Lutzeyer, Michalis Vazirgiannis, and Henrik Boström. Bounding the expected robustness of graph neural networks subject to node feature attacks. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024.arxiv.org/abs/2404.17947
  66. [66]Petar Velikovi, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. In International Conference on Learning Representations, 2018.
  67. [67]Thomas Verelst and Tinne Tuytelaars. Dynamic convolutions: Exploiting spatial sparsity for faster inference. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.DOI
  68. [68]Yoshua Bengio, Nicholas Léonard, and Aaron Courville. Estimating or propagating gradients through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432, 2013.arxiv.org/abs/1308.3432
  69. [69]Mikko I Malinen and Pasi Fränti. Balanced k-means for clustering. In Structural, Syntactic, and Statistical Pattern Recognition: Joint IAPR International Workshop, S+SSPR 2014, Joensuu, Finland, August 20-22, 2014. Proceedings, pages 32–41. Springer, 2014.DOI
  70. [70]Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022.arxiv.org/abs/2207.12598
  71. [71]Cheng Zheng, Bo Zong, Wei Cheng, Dongjin Song, Jingchao Ni, Wenchao Yu, Haifeng Chen, and Wei Wang. Robust graph representation learning via neural sparsification. In International Conference on Machine Learning, pages 11458–11468. PMLR, 2020.
  72. [72]Subhajit Dutta Chowdhury, Zhiyu Ni, Qingyuan Peng, Souvik Kundu, and Pierluigi Nuzzo. Unveiling adversarially robust graph lottery tickets. Transactions on Machine Learning Research, 2024.
  73. [73]Pranjal Awasthi, Himanshu Jain, Ankit Singh Rawat, and Aravindan Vijayaraghavan. Adversarial robustness via robust low rank representations. Advances in Neural Information Processing Systems, 33:11391–11403, 2020.arxiv.org/abs/2007.06555
  74. [74]Pranjal Awasthi, Vaggos Chatziafratis, Xue Chen, and Aravindan Vijayaraghavan. Adversarially robust low dimensional representations. In Conference on Learning Theory, pages 237–325. PMLR, 2021.
  75. [75]Nanqing Xu, Weiwei Feng, Tianzhu Zhang, and Yongdong Zhang. Fd-gan: Generalizable and robust forgery detection via generative adversarial networks. International Journal of Computer Vision, 132(12):5801–5819, 2024.DOI
  76. [76]Dayana Savostianova, Emanuele Zangrando, Gianluca Ceruti, and Francesco Tudisco. Robust low-rank training via approximate orthonormal constraints. Advances in Neural Information Processing Systems, 36:66064–66083, 2023.arxiv.org/abs/2306.01485
  77. [77]Zhebin Wu, Lin Shu, Ziyue Xu, Yaomin Chang, Chuan Chen, and Zibin Zheng. Robust tensor graph convolutional networks via t-svd based graph augmentation. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining, pages 2090–2099, 2022.DOI
  78. [78]Chenhui Deng, Xinyu Li, Zhuo Feng, and Zhiru Zhang. Garnet: Reduced-rank topology learning for robust and scalable graph neural networks. In Learning on Graphs Conference, pages 3–1. PMLR, 2022.arxiv.org/abs/2201.12741
  79. [79]Chenyu Zhou, Wei Huang, Xinyuan Miao, Yabin Peng, Xianglong Kong, Yi Cao, and Xi Chen. Fortifying graph neural networks against adversarial attacks via ensemble learning. Knowledge-Based Systems, 309:112867, 2025.DOI
  80. [80]Jianfu Zhang, Yan Hong, Dawei Cheng, Liqing Zhang, and Qibin Zhao. Defending adversarial attacks in graph neural networks via tensor enhancement. Pattern Recognition, 158:110954, 2025.DOI
  81. [81]Yingzhen Yang. A new concentration inequality for sampling without replacement and its application for transductive learning. In Forty-second International Conference on Machine Learning, 2025.
  82. [82]Yingzhen Yang. Improved generalization bounds for transductive learning by transductive local complexity and its applications. 2025.
  83. [83]Yingzhen Yang. Even sharper bounds for transductive learning and its applications. In Advances in Neural Information Processing Systems (NeurIPS), 2026.arxiv.org/abs/2609.28459
  84. [84]Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 10674–10685. IEEE, 2022.arxiv.org/abs/2112.10752
  85. [85]Kaidi Xu, Hongge Chen, Sijia Liu, Pin-Yu Chen, Tsui-Wei Weng, Mingyi Hong, and Xue Lin. Topology attack and defense for graph neural networks: An optimization perspective. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019, pages 3961–3967. ijcai.org, 2019.arxiv.org/abs/1906.04214
  86. [86]Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Gallagher, and Tina Eliassi-Rad. Collective classification in network data. AI Mag., 29(3):93–106, 2008.
  87. [87]Lada A. Adamic and Natalie Glance. The political blogosphere and the 2004 us election: divided they blog. In Proceedings of the 3rd international workshop on Link discovery, pages 36–43, 2005.
  88. [88]Chenyang Qiu, Guoshun Nan, Tianyu Xiong, Wendi Deng, Di Wang, Zhiyang Teng, Lijuan Sun, Qimei Cui, and Xiaofeng Tao. Refining latent homophilic structures over heterophilic graphs for robust graph convolution networks. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 8930–8938, 2024.arxiv.org/abs/2312.16418
  89. [89]Hongbin Pei, Bingzhe Wei, Kevin Chen-Chuan Chang, Yu Lei, and Bo Yang. Geom-gcn: Geometric graph convolutional networks. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net, 2020.arxiv.org/abs/2002.05287
  90. [90]Jie Tang, Jimeng Sun, Chi Wang, and Zi Yang. Social influence analysis in large-scale networks. In Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 807–816. ACM, 2009.DOI
  91. [91]Felix Mujkanovic, Simon Geisler, Stephan Günnemann, and Aleksandar Bojchevski. Are defenses for graph neural networks robust? Advances in Neural Information Processing Systems, 35:8954–8968, 2022.arxiv.org/abs/2301.13694
  92. [92]Dingyuan Zhu, Ziwei Zhang, Peng Cui, and Wenwu Zhu. Robust graph convolutional networks against adversarial attacks. In KDD, 2019.DOI
  93. [93]Boyuan Feng, Yuke Wang, and Yufei Ding. Uag: Uncertainty-aware attention graph neural network for defending adversarial attacks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 7404–7412, 2021.arxiv.org/abs/2009.10235
  94. [94]Kuan Li, Yang Liu, Xiang Ao, Jianfeng Chi, Jinghua Feng, Hao Yang, and Qing He. Reliable representations make a stronger defender: Unsupervised structure refinement for robust gnn. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining, pages 925–935, 2022.DOI
  95. [95]Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale GAN training for high fidelity natural image synthesis. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019.arxiv.org/abs/1809.11096
  96. [96]Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016.arxiv.org/abs/1512.00567
  97. [97]DC Dowson and BV666017 Landau. The fréchet distance between multivariate normal distributions. Journal of multivariate analysis, 1982.DOI
  98. [98]Zhaocheng Zhu, Zuobai Zhang, Louis-Pascal Xhonneux, and Jian Tang. Neural bellman-ford networks: A general graph neural network framework for link prediction. Advances in Neural Information Processing Systems, 2021.arxiv.org/abs/2106.06935
  99. [99]William L. Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V. N. Vishwanathan, and Roman Garnett, editors, Advances in Neural Information Processing Systems 30, NeurIPS 2017, December 4-9, 2017, pages 1024–1034, 2017.arxiv.org/abs/1706.02216
  100. [100]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. In NeurIPS, 2020.arxiv.org/abs/2005.00687
  101. [101]Kuan Li, Yang Liu, Xiang Ao, and Qing He. Revisiting graph adversarial attack and defense from a data distribution perspective. In The Eleventh International Conference on Learning Representations, 2022.

Paper details

vlnc:W_8k8jwu7d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents