Bias-generating Agent-based-Simulation and its Application to Election Systems

Abstract

Agent-based-Simulation can be used to study the impact of individual decisions on the overalloutcome. There has been a lot of research focused on giving the agent the ability to learn, especially thereare studies that analyze corporate organizations using reinforcement learning. We focus on the electionphenomena that there are often cases where popular candidates are suddenly overtaken by others in theelection campaign. We try to use the characteristic of overfitting of neural networks and conditional reflexlearning, such as the "Pavlov's dog" effect, to explain the phenomena.