Individual and social contributions to learning of mice in Intellicages: the general framework and mean field theory

Abstract

Reinforcement learning enables the adaptation of behaviors to changes in the environment. Models of reinforcement learning based on using previous experience to guide future choices enable balancing exploitation of opportunities and exploring the options and provide accurate predictions of animal behavior in simple instrumental tasks. However, they fail to account for the critical advantage we and other animals have, namely the ability not only to learn from our own choices but also from observing the outcomes of choices made by others. Here, we propose a new conceptual, analytical, and computational framework combining point processes with reinforcement learning models for the description, analysis and modeling of individual and group effects of mice learning reward placement. We show that marked point processes provide a natural language to describe behavior of group housed mice. We show how different reinforcement models of the behavior, including group effects, can be studied effectively in this framework, using example experiments. With this framework we show that the group effects (peer pressure) are twice stronger than individual learning.