Like Miss Marple in an Agathie Christie detective novel, the brain is often faced with the task of inferring a state of the world from noisy and ambiguous clues. An important question is whether, and if so how, the brain performs such inference in a near-optimal manner. This project combines visual decision-making experiments in humans and monkeys with computational modeling and state-of-the-art population recordings in monkey to investigate this question. The project will lead to new insights into the neural code and the relation between neurons and behavior.
A task is used in which an observer classifies a briefly flashed oriented stimulus into one of two classes. The classes are defined by fixed, overlapping probability distributions over orientation. Two common forms of uncertainty play a role in this task. Noise in the sensory observation causes sensory uncertainty, but even if this noise were taken away, a given observation would be consistent with either class; this is an example of ambiguity or class uncertainty. The optimal decision strategy requires the observer to keep track of sensory uncertainty on every trial and to appropriately combine this information with knowledge of the two classes.
The first part of this project will determine whether the computational strategy taken by humans and monkeys during this task is near-optimal. The second part is concerned with whether and how sensory uncertainty is encoded on a trial-by-trial basis in monkey primary visual cortex, and subsequently used in the monkey's decision. This study will put to the test well-known theoretical frameworks for the representation of sensory uncertainty. In the third part, the propagation of sensory uncertainty information from visual cortex to a higher-level decision area, prefrontal cortex, will be examined. Taken together, this project will constitute the first comprehensive test of optimal inference at the level of cortical populations.