A New Network Architecture for Supervised Learning in the Cerebellum
Javier F Medina, Baylor College of Medicine
The cerebellum is often described as a neural machine for supervised learning. Its network architecture consists of anatomically segregated learning modules, which are thought to be specialized for distinct functions defined by the error-related information received via the climbing fiber input, and by the few individual muscles each module is able to control. My talk will present our recent work on mouse eyeblink conditioning, focusing on two unpublished experiments that challenge this classic view about the organization of the cerebellum. First, I will describe a new recurrent circuit that allows some Purkinje cells in the cerebellar cortex to learn in the absence of error-related information in their climbing fiber input. Second, I will show that the output of a single cerebellar module can be used to control a complex motor synergy that requires coordination of multiple muscles. Altogether, the results suggest a new organizational framework for understanding what the cerebellum learns, and how.