One of the most impressive feats the brain performs is its use of information that is rife with uncertainty to successfully guide behavior. That the world is full of ambiguous stimuli and perceived through noisy sensors is clear. However, whether and how the brain makes use of this uncertainty is an open and important question. Behavioral studies using tasks with noisy or ambiguous stimuli suggest that subjects' performance is nearly optimal. These results strongly suggest that the brain must represent and use information about sensory uncertainty itself in addition to information about parameters of the stimulus, yet many theories about neural coding do not account for how such uncertainty could be represented or used. The theory of Probabilistic Population Coding (PPC) does provide such an account, claiming that populations of neurons encode estimates of stimulus parameters and uncertainty regarding those estimates simultaneously in the form of a ?likelihood function? over the stimulus. PPCs further provide neurally plausible mechanisms for performing a number of Bayesian computations, however, this promising theory has never been tested neurophysiologically at the population level. To study the mechanisms by which the brain represents and uses sensory uncertainty to guide adaptive behavior, the current proposal will combine multi-electrode recordings, computational neuroscience and psychophysics. Specifically, we will study visual decision-making in an orientation classification task, which we have previously shown that primates perform near optimally and which requires the use of uncertainty information to achieve optimal performance.
In Aim 1, we will test the hypothesis that populations of neurons jointly encode likelihood functions as predicted by PPC, by recording from V1 while subjects perform the classification task.
In Aim 2, we will test whether a shared encoding of uncertainty information between V1 and prefrontal cortex (PFC) leads to a functional correlation between these two areas. Through our combination of in vivo population recordings in awake, behaving primates, the proposed project is strongly positioned to test the core hypothesis behind PPC and elucidate the mechanisms by which the brain makes possible optimal behavioral performance in a noisy environment.
Decision-making is difficult because our sensory observations are typically corrupted by uncertainty, and in order to perform optimally during decision-making, we must take into account the current level of sensory uncertainty. Using state of the art neuronal population recordings, carefully crafted behavioral task and computational methods we will study how uncertainty is represented in the brain, directly testing a leading framework for sensory information encoding for the first time. This project will serve as a critical step in understanding the nature of the neural code and provide a basis for improved decoding of information from the brain that promises to eventually allow for the development of more efficient use of brain machine interfaces.