Identifying ADHD-Related Abnormal Functional Connectivity with a Graph Convolutional Neural Network

Figure 2

A 2-layer GCN model with a self-attentive pooling mechanism was built in this study and called 2L_AGCN. The number of convolutional kernels is 32 and 16; the pooling parameter is 0.35; the activation function is ReLU; the learning rate is 0.034; and the weight decay is 0.00005.