Lixiang Wu, UT-Austin

"Developing Language Encoding Models"

Abstract: As the fast development of brain imaging technology and computational methods, researches on human brain functions draw more attention. Our work focuses on studying how the human brain processes natural language with fMRI data and computational models. I'll introduce the physics of fMRI, and our previous work: language encoding model, which predicts human brain response from language stimuli and maps semantic selectivity across the cortex. Then I'll explain the limitations of the current language encoding model, and talk about how we improve the model performance by aligning human brain data across multiple stimuli and multiple subjects with shared response model, which is called the sparse experimental design. Furthermore, I want to discuss my future plans on interpreting semantic map on human cortex as spatial gradients, and fusing fMRI data with EEG data to study brain functions at high spatial-temporal resolution.