Abstract
The brain creates internal representations of the external world in the form of neural activity, which is structured to support adaptive behavior. In many cortical regions, individual neurons respond to specific features that are matched to the function of each region and statistics of the world. In frontal cortex, however, neurons display baffling complexity, responding to a mixture of sensory, motor and other variables. Here we use an integrated new approach to understanding the architecture of higher-order cortical representations, and use this approach to show that discrete groups of orbitofrontal cortex (OFC) neurons encode distinct decision variables. Using rats engaged in a complex task combining perceptual and value guided decisions, we found that OFC neurons can be grouped into distinct, categorical response types. These categorical representations map directly onto decision-variables of a choice model explaining our behavioral data, such as reward size, decision confidence and integrated value. We propose that, like sensory neurons, frontal neurons form a sparse and over complete population representation aligned to the natural statistics of the world – in this case spanning the space of decision-variables required for optimal behavior.
