Abstract

At the algorithmic level, neural computations can be described by the transmission, manipulation, and interaction of multiple distinct representations within and across brain regions. In visual neuroscience, decades of research has uncovered a hierarchy of successive and recurrent abstractions of incoming visual information from simple and complex coding in early visual cortex to sparse modality-independent “concept cells” in the temporal lobe. In this talk, I will present work aimed at developing an analogous understanding in the domain of decision-making and control via the application of general linear modeling (based on behavioral models), multivoxel pattern decoding, and connectivity analysis to functional magnetic resonance imaging data. Through combinations of these techniques, nontrivial intermediate representations and hierarchical modes of neural processing will be shown to be utilized by the brain in the production of rational behavior. Together, the results of these studies constrain the set of possible algorithms being implemented by the brain in decision-making, action selection, and abstract inference.