Browse Teach Curate

Long-Range Dependency Capture in Brain Graph Transformers via Biased Random Walks

A brain graph transformer (ALTER) that models functional-connectivity networks from fMRI and captures long-range dependencies between regions of interest using a biased random walk whose transition probabilities are weighted by inter-ROI Pearson correlation. The resulting per-node long-range embeddings are concatenated with node features and integrated with short-range structure through a self-attention encoder for neurological disease classification.

Data Text

Long-range Brain Graph Transformer Shuo Yu1,2, Shan Jin3, Ming Li4,5, Tabinda Sarwar6, Feng Xia6B

This paper introduces ALTER (Adaptive Long-range aware TransformER), a graph transformer for analyzing brain functional-connectivity networks derived from fMRI, where nodes are regions of interest (R…