The world's complexity makes sensory information ambiguous. A set of signals sweeping across the retina, for instance, might be generated by a moving object or by the animal's own motion. The brain resolves this ambiguity by constructing an internal model of reality that associates patterns in sensory data with events in the world, in a process called causal inference. To support adaptive action, the brain's estimates of all relevant variables must be updated to remain consistent with physics under the current interpretation of the scene, as well as with each other. This proposal focuses on perceptual interactions among object motion, self-motion, and depth. This project will test the predictions of models from Project A for how causal inference should update these related sensory variables. Based on previous work and preliminary data, the overall hypothesis is that parietal (7a) and prefrontal (8av) areas signal whether or not an object moves in the world, and that these signals flow through feedback projections to update sensory representations of object depth in area MT and self-motion velocity in area MSTd. The goal of this project is to use traditional, trial-based tasks to determine whether, and if so how, causal inferences are propagated back to sensory cortex to update representations of task-relevant variables in monkeys. The goal of Aim 1 is to test whether causal inference modulates low-level sensory representations of motion by examining neural correlates of optic flow parsing, a phenomenon in which the perceived direction of an object's motion is strongly and predictably influenced by background optic flow.
Aim 2 will test directly if sensory representations of task variables are updated to maintain consistency with beliefs about the world. Monkeys will judge whether an object moves in the world and also report its depth, while neural populations in parietal area 7a, prefrontal area 8av, and sensory areas MT and MSTd are recorded. The theoretical framework, supported by preliminary behavioral data, predicts that this causal inference about object motion will induce specific patterns of bias in estimates of depth and self-motion velocity, and that neural estimates of motion and depth in MT and MSTd will update to remain consistent with behavior. To identify the neural circuitry necessary for causal inference and sensory updating, Aim 3 will inactivate feedback pathways with muscimol or optogenetics while neural activity is recorded in sensory representations, as animals perform the same task as in Aim 2.
Aim 1 is expected to establish that causal inference modulates early visual processing, and to identify the areas where these effects are implemented.
Aim 2 will provide the first cellular and neural population evidence of causal inference and sensory updating by belief propagation.
Aim 3 will establish a neural circuit that is necessary for mediating causal inference and for updating sensory representations. Together, successful completion of the proposed experiments will uncover circuit mechanisms of causal inference and sensory variable updating and establish a new functional role for top-down feedback connections, which are a general feature of cortical organization.