Parallel Hypergraph Convolutional Neural Networks for Image Annotation

Wang, Mengke; Liu, Weifeng; Yuan, Xin’an; Li, Wei; Liu, Baodi

doi:10.23919/ccc55666.2022.9901938

Cited by 1 publication

(13 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where X E ∈ R m×d denotes the hyperedge features, and f V→E (⋅) and f E→V (⋅) are two feature aggregation functions. More recent developments include models by Wang et al(2023a) and Wang et al(2023b). Wang et al(2023a) design the model from a hypergraph diffusion perspective, while Wang et al(2023b) unroll an optimisation algorithm minimising an hypergraph energy function.…”

Section: Gnns and Hypergnnsmentioning

confidence: 99%

“…More recent developments include models by Wang et al(2023a) and Wang et al(2023b). Wang et al(2023a) design the model from a hypergraph diffusion perspective, while Wang et al(2023b) unroll an optimisation algorithm minimising an hypergraph energy function. In the following, we concentrate on five state-of-the-art HyperGNNs: UniGNN , AllDeepSets (Chien et al, 2022), AllSetTransformer (Chien et al, 2022), ED-HNN (Wang et al, 2023a), and PhenomNN (Wang et al, 2023b).…”

Section: Gnns and Hypergnnsmentioning

confidence: 99%

“…With the eigenvectors derived from the specially designed matrix, the hypergraph node classification task can be solved in two steps: 1) The eigenvectors are utilised as inputs by a clustering algorithm that separates nodes into distinct clusters; and 2) Unlabeled nodes in each cluster are assigned labels based on the most common labels in the cluster. Most deep-learning-based methods (Feng et al, 2019;Bai et al, 2021;Chien et al, 2022;Wang et al, 2023a;b) develop HyperGNNs through adjusting the designs of existing GNNs (Kipf & Welling, 2017;Bodnar et al, 2022;Atwood & Towsley, 2016;Gilmer et al, 2017;Xu et al, 2019;Chen et al, 2021;Yang et al, 2021) to classify nodes in an end-to-end paradigm.…”

Section: Related Workmentioning

confidence: 99%

“…Datasets. We use nine publicly available datasets, including five homophilic hypergraphs (Cora, Citeseer, Pubmed, Cora-CA, and DBLP-CA) from Yadati et al (2019), and four heterophilic hypergraphs (Congress, Senate, Walmart and House) from Wang et al (2023a). The details of these datasets are summarised in Table 2.…”

Section: Experiments Setupmentioning

confidence: 99%

“…The success of graph neural networks (GNNs) (Wu et al, 2020) highlights the proficiency of neural networks in processing data with pairwise interactions. Inspired by this, many hypergraph neural networks (HyperGNNs) have been proposed and widely used in the hypergraph node classification task (Feng et al, 2019;Bai et al, 2021;Chien et al, 2022;Wang et al, 2023a;b;Duta et al, 2023). These models are typically based on designs specific to hypergraphs, such as operators for learning hyperedge features Chien et al, 2022) or hypergraph energy functions (Wang et al, 2023b).…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Graph Neural Networks for Node Classification

Tang

Liao

2022

Graph Neural Networks: Foundations, Frontiers, and Applications

View full text Add to dashboard Cite

This work analyzes Graph Neural Networks, a generalization of Fully-Connected Deep Neural Nets on Graph structured data, when their width, that is the number of nodes in each fullyconnected layer is increasing to infinity. Infinite Width Neural Networks are connecting Deep Learning to Gaussian Processes and Kernels, both Machine Learning Frameworks with long traditions and extensive theoretical foundations. Gaussian Processes and Kernels have much less hyperparameters then Neural Networks and can be used for uncertainty estimation, making them more user friendly for applications. This works extends the increasing amount of research connecting Gaussian Processes and Kernels to Neural Networks. The Kernel and Gaussian Process closed forms are derived for a variety of architectures, namely the standard Graph Neural Network, the Graph Neural Network with Skip-Concatenate Connections and the Graph Attention Neural Network. All architectures are evaluated on a variety of datasets on the task of transductive Node Regression and Classification. Additionally, a Spectral Sparsification method known as Effective Resistance is used to improve runtime and memory requirements. Extending the setting to inductive graph learning tasks (Graph Regression/ Classification) is straightforward and is briefly discussed in 3.5.

show abstract