Reproduce “Self-supervised Graph Transformer with Contrastive Learning for Brain Connectivity Analysis towards Improving Autism Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yicheng Leng, Syed Muhammad Anwar, Islem Rekik, Sen He, Eung-Joo Lee
Researchers developed a self-supervised graph transformer model with contrastive learning and a specialized graph augmentation strategy to enhance autism detection using fMRI-derived brain connectivity data. The method achieved an AUROC of 82.6 1.8 and an accuracy of 74.4 2.4 on the ABIDE dataset, surpassing previous GNN and Graph Transformer approaches.
Key insights
- Self-supervised learning with contrastive objectives effectively addresses data scarcity by enabling models to learn robust representations from unlabeled fMRI data.
- Customized graph augmentation strategies, like dilation and shrinkage tailored for dense correlation matrices, are essential for improving feature learning in brain network analysis.
- Integrating a Graph Transformer architecture with CSSL and specific noise addition significantly improves both the performance and stability of models for neurological disorder detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.