Predicting Decision-Making: An Updated Computational Analysis on Reinforcement Learning.

Abstract

Trying to map the behavior of decision-making is never simple. Specifically, how does one begin to assign the probability of an action occurring properly in reward and reinforcement learning? Decision-making is a part of every single day. Some actions require more thought about past experiences and future rewards and consequences. Learning how to model this process is incredibly complex. In the future, these types of analysis apply to regular decision-making and those associated with neurological disorders. Understanding each step of the decision-making process regarding disorders such as addiction or OCD is a vital step in learning better ways to intervene and treat those affected. In this thesis, a computational analysis of rodents in a two-step task will be conducted when certain brain regions are inactivated. The result of this paper creates updated modeling parameters for specific agents within reinforcement learning. Each agent attributed to model-based and model-free is mapped between brain regions and all rats used in the study. The updated model creates a more accurate narrative of the decision-making in reinforcement learning.