2021-04-12l 조회수 3330
Neural computations underlying Bayesian timing behaviors
Statistical regularities in the environment create prior beliefs that we rely on to optimize our behavior when sensory information is uncertain. For example, a baseball hitter not only measures speed of the incoming ball but also uses prior knowledge about the pitcher to prepare movements. Bayesian theory formalizes how prior beliefs can be leveraged and has had a major impact on models of perception, sensorimotor function, and cognition. However, it is not known how the brain implements Bayesian computations. Using time perception as a case study, I tackle this question by recording neural population activity of dorsomedial frontal cortex in non-human primates. We found a neural mechanism for encoding Bayesian time estimates through warping of neural population dynamics in the dorsomedial frontal cortex. This mechanism also emerged in a task-optimized recurrent neural network, suggesting its generality as a computational motif.
In the second part of the talk, I will describe an ongoing work dissecting the laminar structure of the cortical population activity. Based on recent work implicating higher-order thalamus in the contextual modulation of cortical dynamics, we hypothesized that the prior-dependent modulation of population activity in the frontal cortex may originate in the superficial layers where thalamocortical projections terminate. To test this hypothesis, we analyzed neural activities recorded simultaneously from neurons across the cortical laminae. Our initial analyses indicated that the response profile of individual neurons and the dimensionality of the population activity were similar across laminae. However, we found that the context signals were initially present and subsequently amplified in the superficial – not deep – layers. These results suggest that the laminar organization of cortical microcircuits may provide a scaffold for further dissecting how cortical dynamics enable behaviorally-relevant computation.