BYU

Abstract by Alexia Delorey

Personal Infomation


Presenter's Name

Alexia Delorey

Degree Level

Undergraduate

Abstract Infomation


Department

Computer Science

Faculty Advisor

David Wingate

Title

Basics of Deep RL

Abstract

My research is focused on reinforcement learning and applications to neural networks. Reinforcement learning focuses on an agent, or the thing that is doing the learning. My research specifically focused on the applications of the Q-learning algorithm on helping an agent to find the best path through a maze. Inside the maze, the agent must first decide whether to be greedy or exploratory. Once it has decided that it then moves into a new state, which was either chosen from the Q matrix or at random. This state has a reward. This reward is given to the Q algorithm, which calculates the Quality of that action for that state based on both this instance and previous experience. This value is then stored in a Q matrix. The agent continues to run through the maze, learning from its mistakes and updating the Q matrix until it knows the best path through the maze. This research has far reaching applications to the world of Deep RL. It opens up the possibility of teaching computers to learn for themselves, based on their own experience, rather than being told what to do for every state they may find themselves in.