Reinforcement learning of the neocortico-hippocampal system

Hypothetical neural mechanisms of how the animal learns to obtain reward by trial and error are presented. To find out which trial was rewarded in what state of environment, two kinds of memory are necessary: (1) short-term memory temporally remembering trials and environmental states in which they are made and (2) long-term memory storing pairs of the trials and states reinforced by reward. As neural entities of the two memories, long-term potentiation of hippocampus and motor cortex and cortical plasticity regulated by the hypothalamic reward system are assumed. Physiological and anatomical data concerning them are integrated into a set of hypotheses describing how the real nervous system performs the learning using the neuroplasticities. A mathematical model of the nervous system is developed where single neuron behavior is given by the Hodgkin-Huxley electric circuit. Computer simulation of the model behavior is conducted to test the hypotheses.