Collecting reward to defend homeostasis: A homeostatic reinforcement learning theory

Efficient regulation of internal homeostasis and defending it against perturbations requires complex behavioral strategies. However, the computational principles mediating brains homeostatic regulation of reward and associative learning remain undefined. Here we use a definition of primary rewards, as outcomes fulfilling physiological needs, to build a normative theory showing how learning motivated behavior is modulated by the internal state of the animal. The theory proves that seeking rewards is equivalent to the fundamental objective of physiological stability, defining the notion of physiological rationality of behavior. We further give a formal basis for temporal discounting of reward. It also explains how animals learn to act predictively to preclude prospective homeostatic challenges, and attributes a normative computational role to the modulation of midbrain dopaminergic activity by hypothalamic signals.