Towards data augmentation in graph neural network: An overview and evaluation

Adjeisah, Michael; Zhu, Xinzhong; Xu, Huiying; Ayall, Tewodros Alemu

doi:10.1016/j.cosrev.2022.100527

Cited by 10 publications

(3 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, interactions between cells along spatial gradients or domain boundaries might also be a potential angle for graph augmentation. Some other studies also suggest to seek for cells shared similar biological characterizations or annotations (domains) or functional modules from the same datasets or from replicates 55, 56 . In addition, when the spatial datasets have a temporal series, creating positive views which capture similar patterns or dynamic changes over times might be another effective method for graph augmentation.…”

Section: Discussionmentioning

confidence: 99%

Spatial domains identification in spatial transcriptomics by domain knowledge-aware and subspace-enhanced graph contrastive learning

Gui,

Li,

2024

Preprint

View full text Add to dashboard Cite

Spatial transcriptomics (ST) technologies have emerged as an effective tool to identify the spatial architecture of the tissue, facilitating a comprehensive understanding of organ function and tissue microenvironment. Spatial domain identification is the first and most critical step in ST data analysis, which requires thoughtful utilization of tissue microenvironment and morphological priors. To this end, we propose a graph contrastive learning framework, GRAS4T, which combines contrastive learning and subspace module to accurately distinguish different spatial domains by capturing tissue microenvironment through self-expressiveness of spots within the same domain. To uncover the pertinent features for spatial domain identification, GRAS4T employs a graph augmentation based on histological images prior, preserving information crucial for the clustering task. Experimental results on 8 ST datasets from 5 different platforms show that GRAS4T outperforms five state-of-the-art competing methods in spatial domain identification. Significantly, GRAS4T excels at separating distinct tissue structures and unveiling more detailed spatial domains. GRAS4T combines the advantages of subspace analysis and graph representation learning with extensibility, making it an ideal framework for ST domain identification.

show abstract

Section: Discussionmentioning

confidence: 99%

Spatial domains identification in spatial transcriptomics by domain knowledge-aware and subspace-enhanced graph contrastive learning

Gui,

Li,

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…An effective method for mitigating this problem is to perform data augmentation by manipulating the existing data and generating synthetic information [1][2][3][4][5][6][7][8]. Existing methods can be categorized according to the types of manipulated graph elements, that is, node attributes, graph structure, and node labels [3,[9][10][11][12]. The proposed GMMDA addresses all these elements by generating synthetic nodes.…”

Section: Related Workmentioning

confidence: 99%

“…Recently, graph data augmentation (GDA) has been frequently studied as an effective technique for enhancing the generalization of graph neural networks (GNNs) for semisupervised node classification [1][2][3][4][5][6][7][8]. Augmentation can be realized by manipulating one or more types of graph elements, that is, node attributes, graph structure, and node labels, as summarized in several surveys [3,[9][10][11][12]. Thus, GDA can bring plausible variations to a given graph, thereby eliminating the over-smoothing (i.e., mixing of nodes from different classes) [13][14][15][16] and overfitting of GNNs, particularly when labeled data are limited.…”

Section: Introductionmentioning

confidence: 99%

GMMDA: Gaussian Mixture Modeling of Graph in Latent Space for Graph Data Augmentation

Li,

Xu,

Yamanish

2024

Preprint

View full text Add to dashboard Cite

Graph data augmentation (GDA), which manipulates graph structure and/or attributes, has been demonstrated as an effective method for improving the generalization of graph neural networks on semi-supervised node classification. As a data augmentation technique, label-preservation is critical, that is, node labels should not change after data manipulation. However, most existing methods overlook the label-preservation requirements. Determining the label-preserving nature of a GDA method is highly challenging, owing to the non-Euclidean nature of the graph structure. In this study, for the first time, we formulate a label-preserving problem (LPP) in the context of GDA. The LPP is formulated as an optimization problem in which, given a fixed augmentation budget, the objective is to find an augmented graph with minimal difference in data distribution compared to the original graph. To solve the LPP problem, we propose GMMDA, a generative data augmentation (DA) method based on Gaussian mixture modeling (GMM) of a graph in a latent space. We designed a novel learning objective that jointly learns a low-dimensional graph representation and estimates the GMM. The learning is followed by sampling from the GMM, and the samples are converted back to the graph as additional nodes. To uphold label preservation, we designed a minimum description length (MDL)-based method to select a set of samples that produces the minimum shift in the data distribution captured by the GMM. Through experiments, we demonstrate that GMMDA can improve the performance of graph convolutional network on Cora, Citeseer and Pubmed by as much as 7.75%, 8.75%, and 5.87%, respectively, significantly outperforming the state-of-the-art methods.

show abstract

GMMDA: Gaussian mixture modeling of graph in latent space for graph data augmentation

Li,

Xu,

Yamanishi

2024

Knowl Inf Syst

View full text Add to dashboard Cite

Graph data augmentation (GDA), which manipulates graph structure and/or attributes, has been demonstrated as an effective method for improving the generalization of graph neural networks on semi-supervised node classification. As a data augmentation technique, label preservation is critical, that is, node labels should not change after data manipulation. However, most existing methods overlook the label preservation requirements. Determining the label-preserving nature of a GDA method is highly challenging, owing to the non-Euclidean nature of the graph structure. In this study, for the first time, we formulate a label-preserving problem (LPP) in the context of GDA. The LPP is formulated as an optimization problem in which, given a fixed augmentation budget, the objective is to find an augmented graph with minimal difference in data distribution compared to the original graph. To solve the LPP problem, we propose GMMDA, a generative data augmentation (DA) method based on Gaussian mixture modeling (GMM) of a graph in a latent space. We designed a novel learning objective that jointly learns a low-dimensional graph representation and estimates the GMM. The learning is followed by sampling from the GMM, and the samples are converted back to the graph as additional nodes. To uphold label preservation, we designed a minimum description length (MDL)-based method to select a set of samples that produces the minimum shift in the data distribution captured by the GMM. Through experiments, we demonstrate that GMMDA can improve the performance of graph convolutional network on Cora, Citeseer and Pubmed by as much as $$7.75\%$$ 7.75 % , $$8.75\%$$ 8.75 % and $$5.87\%$$ 5.87 % , respectively, significantly outperforming the state-of-the-art methods.

show abstract

Towards data augmentation in graph neural network: An overview and evaluation

Cited by 10 publications

References 8 publications

Spatial domains identification in spatial transcriptomics by domain knowledge-aware and subspace-enhanced graph contrastive learning

Spatial domains identification in spatial transcriptomics by domain knowledge-aware and subspace-enhanced graph contrastive learning

GMMDA: Gaussian Mixture Modeling of Graph in Latent Space for Graph Data Augmentation

GMMDA: Gaussian mixture modeling of graph in latent space for graph data augmentation

Contact Info

Product

Resources

About