Metalearning and neuromodulation.

Michael E Hasselmo

Michael E Hasselmo

Michael E Hasselmo

+ -1 Evaluators

This important work presents the hypothesis that various neuromodulatory systems could be understood in the framework of parameters used in a class of neural network models referred to as reinforcement learning models. Reinforcement learning models present an excellent theoretical framework for understanding the ongoing interactions of an agent seeking reward in an environment. There is already extensive evidence supporting a role for dopamine as an error signal in reward prediction, and the proposed role of acetylcholine as a learning rate is consistent with physiological data and other models. The proposals about norepinephrine and serotonin are more speculative, but provide an interesting and well-developed theoretical hypothesis in a field which needs more models which could link the behavioral role of modulators to specific physiological effects of these modulators. Reinforcement learning models have already been mapped to elements of basal ganglia circuitry. Further mapping of these models to other brain structures will allow mapping of the theoretical roles proposed here to specific physiological cellular processes.