Abstract:
This abstract describes a reinforcement learning project that is meant to compare the effects of various variables and architectures on learning and artificial intelligence. Results will be quantified in its convergence speed, resource usage, etc. In addition, results will be analyzed statistically to look for patterns and determine favorable outcomes that are widely applicable to other unrelated reinforcement learning agents. Reinforcement learning in terms of this paper consists of appropriate setup of input features and output results, configuration of the architecture of the network, appropriate reward shaping, and proper sampling of data and testing advancements with proper analytics. The hardware requirements of this project are met with cost-effective workstation platforms applied in a typical laboratory setting. Software requirements are all free and able to be used in an education or personal setting, the specifics of which will be discussed in this paper. The project is made up of a reinforcement self-learning agent, the use of present game software, a custom program to create input, and a model to interpret our results. The environment of choice will be games simple enough for an introductory student to develop a reinforcement learning model for. These video game(s) will be used as a way to pique interest and increase the connection to the material and relatability to the results. In addition to serving as a study into the behavior and nature of networks, the project can also serve as a reference for student research or learning and more advanced projects can be built upon it.
Team Members
Benjamin Lubina (Ramakrishnan Sundaram) – Gannon University
Download the Project Summary