ABSTRACT

This chapter reviews how the basal ganglia (BG) can be reorganized through learning. In reinforcement learning, some action occurs, and the reinforcement signal changes the value function. The corresponding synaptic weights do not change until reinforcement. In the control hierarchy, learning is only needed when the existing organization is inadequate. In reorganization, parameter variation is initiated by persistent errors in essential controlled variables and manifested in exploration and behavioral variability. Error reduction is responsible for stopping the variation and saving the parameters. The study of bird song has provided insights into behavioral variability and reorganization in learning. During song learning, birds must first establish an internal template of the tutor song and an internal comparison function that compares the template with the actual song performance. They can vary their outputs until the input matches the template. The BG not only contribute to online correction during performance but are also critical for discovering and preserving the right system parameters. As adaptive gain, dopamine (DA) can repeat specific action variants that reduce error. By repeating the right parameters while the error remains high, DA can contribute to the induction of long-term plasticity.