Measuring and Modeling Transformations of Information Between Brain Regions with fMRI

Investigating how information is transformed from brain region to brain region is a crucial step to understand the neural foundations of cognitive processes. This investigation requires a characterization of the representations encoded in different regions, and models of how they are transformed that can match the complexity of neural processes. We introduce an approach in which representations are characterized as points in multidimensional spaces, and processes transforming representations from region to region are modeled as nonlinear functions using artificial neural networks. Across multiple experiments with different stimuli and tasks, we show that this approach reveals functionally relevant network structure and outperforms comparable linear models at predicting independent data.