Customized Graph Attention Layer for Graph Neural Network
Customized Graph Attention Layer based on Deep Learning Toolbox for graph neural network.
This fileexchange provides a native Deep Learning Toolbox custom implementation of the Graph Attention Layer (GAT) refers to the Multilabel Graph Classification Using Graph Attention Networks. It supports multi-head masked self-attention, dynamic neighbor weight assignment, standard attention normalization and full integration with MATLAB’s dlnetwork framework.
It should be noted that the inputs of this layer are features and adjacency matrix with format "SCB" and "SSB", respectively. The output of this layer has format "SCB", where "C"=outSize when the aggregation=="cat" and "C"=outSize/numHeads when aggregation=="mean".
引用格式
Chuguang Pan (2026). Customized Graph Attention Layer for Graph Neural Network (https://ww2.mathworks.cn/matlabcentral/fileexchange/184011-customized-graph-attention-layer-for-graph-neural-network), MATLAB Central File Exchange. 检索时间: .
| 版本 | 已发布 | 发行说明 | Action |
|---|---|---|---|
| 1.0.0 |
