Self-supervised learning for representing and decoding brain activity linked to behavior
Add to Google Calendar
Deep learning has been increasingly applied to functional magnetic resonance imaging (fMRI) data for behavior prediction, disease classification, and treatment recommendation. However, such applications face common challenges in generalizability, specificity, and explainability. When models learn to optimize the performance for a specific goal or dataset, they often fail to generalize to other goals or datasets. When models are trained with “big data” from many individuals, they often disregard individual differences. When models act as a “black box”, their inner workings are difficult to explain. Hence, existing deep learning models applied to fMRI are inadequate for mechanistic and individualized understanding of brain-behavior association.
To address these challenges, we propose a scalable, modular, generalizable, and explainable artificial intelligence (XAI) system to represent and decode fMRI activity linked to behavior specific to each individual.
Zoom information will be sent to e-HAIL members.