Fig. 1: The model framework and workflow.
From: Identifying autism spectrum disorder from multi-modal data with privacy-preserving
Multimodal data including fMRI scan data and phenotype data are used to develop local models for ASD diagnostic tasks. a Federated Learning Framework. Each local model trains it using protected private data and communicates with the global model at a specific frequency, uploading only the encrypted model parameters when communicating (left). The global model uses the average strategy to update the model parameters and distribute them to each local model (right). b The local model uses multimodal data to construct hyperedge groups separately, and generate the hypergraph by connecting the hyperedge groups (Hypergraph Generation). The hypergraph and the fused node features are jointly input to the hypergraph neural network for multi-layer hypergraph convolution to finally obtain the classification results.
