Visual learning is critical to the lives of human and non-human primates. Visuomotor association, the assignment of an arbitrary symbol to a particular movement (like a red light to a braking movement), is a well- studied form of visual learning. This proposal tests the hypothesis that the brain accomplishes visuomotor associative learning using an anatomically defined closed-loop network, including the prefrontal cortex, the basal ganglia, and the cerebellum. In our preliminary work we have developed a task that studies how monkeys learn to associate one of two novel fractal symbols with a right hand movement, and the other symbol with a left hand movement. Every experiment begins with the monkeys responding to two overtrained symbols that they have seen hundreds of thousands of times. At an arbitrary time we change the symbols to two fractal symbols that the monkey has never seen. It takes the monkey 40 to 70 trials to learn the new associations. In our preliminary results we have discovered that Purkinje cells in the midlateral cerebellar hemisphere track the monkeys? learning as they as they figure out the required associations. The neurons signal the result of the prior decision. Half of the neurons respond more when the prior decision was correct; the others respond more when the prior decision was wrong. The difference between the activity of these two types of neurons provides a cognitive error signal that is maximal when the monkeys are performing at a chance level, and gradually becomes not different from zero as the monkeys learn the task. The neurons do not predict the result of the impending decision. Although the neurons change their activity dramatically at the symbol switch, the kinematics of the movements do not change at all. This proposal takes this discovery as the starting point for four aims: 1) to use viral transynaptic tract tracing to discover the cortical and basal ganglia regions that project to the cerebellar visuomotor association area. 2) to record from the four nodes of the network as anatomically defined (midlateral cerebellar hemisphere, dentate nucleus, basal ganglia, prefrontal cortex), simultaneously, using multiple single neuron recordings, to see if these areas also have information about the process of visuomotor association 3) to inactivate each node, to see how their inactivation affects the monkey?s ability to learn new associations, and whether the inactivation affects the activity of the neurons at the other nodes. 4) to develop computational methods to analyze the activity of neural activity recorded simultaneously in all four nodes of the network (Aim 2) in the midlateral cerebellar cortex with regard to parameters such as prior outcome and movement, hand, symbol, and the intensity and epoch of the prior cognitive error signal. We will use dimensional reduction techniques to answer questions like whether hand or symbol can be decoded from network activity. We will model how the cerebellum simple spike cognitive error signal might propagate through the network and be used to facilitate visuomotor association learning and the processing of signals in the cerebellum, basal ganglia and cerebral cortex
Learning that a particular object cues a particular action, as a red light makes us stop walking or brake the car, is critical for human behavior and can be degraded by human disease. This project will apply physiological, computational, and anatomical methods to investigate a brain network for visual learning. We will find the exact areas of the cerebral cortex, basal ganglia, and cerebellum that participate in this learning, and use machine learning techniques to understand how the activity of neurons recorded simultaneously in these brain areas can facilitate learning.
