ENS, salle 236, 24 rue Lhomond, 75005 Paris
The links between brain activity and behavior are often studied using very constrained tasks involving simple binary choices. We wanted to operationalize the notion of the use of sensory (specifically auditory) information to drive behavior, and for that purpose we
developed a setup (the RIFF - rat interactive fantasy facility) that allows us to implement general Markov decision processes (MDPs) with a large number of action choices (up to 12 different ports for accessing food and water) based on auditory cues. The use of MDPs makes it possible to apply reinforcement learning theory in order to calculate
optimal policies. Rats can freely move in the RIFF, and brain activity is recorded using telemetry or neural loggers on the animal. I will describe the RIFF and the theory underlying it, and present preliminary results regarding (1) the behavior of rats in the RIFF and its correspondence with optimal policies as well as optimal policies under information constraints; (2) neural activity recorded in primary auditory cortex as well as in the auditory field within the insular cortex, and its dependence on behavioral state.