Learning Robust and Task-Invariant Functional Representation from fMRI through Siamese Self-Supervised Learning

In three linesBrainSimSiam, a lightweight self-supervised learning framework, learns robust representations from fMRI data without labels. Using positive-only pairs, it generalizes across multiple tasks (classification, regression) and outperforms supervised baselines, reducing computational requirements for foundation models in neuroimaging.

Summary generated by Claude — human-verified