Princeton Neuroscience Institute, Princeton University, Department of Psychology
In recent years, ideas from the computational field of reinforcement learning have revolutionized the study of learning in the brain, famously providing new, precise theories about the effects of dopamine on learning in the basal ganglia. However, the first ingredient in any reinforcement learning algorithm is a representation of the task as a sequence of "states". Where do these states come from? In this talk I will first argue, and demonstrate using behavioral experiments, that animals and humans learn the latent structure of a task using Bayesian inference (or an approximation thereof), thus forming a state space through experience. I will interpret these findings in terms of an ongoing battle between learning and memory, whose outcomes are ultimately determined by prediction errors. I will then suggest that the orbitofrontal cortex is critical to representing these learned states, in particular in tasks in which states depend not only on externally observable information but also on internal information, for instance from working memory or in social scenarios.
1. SJ Gershman, KA Norman & Y Niv (2015) – Discovering latent causes in reinforcement learning – Current Opinion in Behavioral Sciences 5:43-50. http://www.princeton.edu/~nivlab/papers/GershmanNormanNiv2015.pdf
2. N Schuck, M Cai, RC Wilson & Y Niv (2016) – Human orbitofrontal cortex represents a cognitive map of state space – Neuron 91, 1402-1412 http://www.princeton.edu/~nivlab/papers/SchuckNiv2016.pdf