A NEURAL-NETWORK MODEL OF THE DYNAMICS OF HUNGER, LEARNING, AND ACTION VIGOR IN MICE

Recently the computational-neuroscience literature on animals' learning has proposed some models for studying organisms' decisions related to the energy to invest in the execution of actions ("vigor"). These models are based on average reinforcement learning algorithms which make it possible to reproduce organisms' behaviours and at the same time to link them to specific brain mechanisms such as phasic and tonic dopamine-based neuromodulation. This paper extends these models by explicitly introducing the dynamics of hunger, driven by energy consumption and food ingestion, and the effects of hunger on perceived reward and, consequently, vigor. The extended model is validated by addressing some experiments carried out with real mice in which reinforcement schedules delivering lower amounts of food can lead to a higher vigor compared to schedules delivering larger amounts of food due to the higher perceived reward caused by higher levels of hunger.