Date

Fall 12-10-2019

Document Type

Honors Project

First Advisor

Zacharski, Ron

Department Chair or Program Director

Finlayson, Ian

Degree Name

Bachelor of Science

Major or Concentration

Computer Science

Department or Program

Computer Science

Abstract

Deep Reinforcement Learning has shown great progress in domains such as the Atari Arcade Learning Environment. The problem, however, is that the agent playing the game requires many interactions before it starts to show good performance. This is okay for some domains such as video games, but as we start to look towards integrating deep reinforcement learning into real world applications, we need to minimize the number of interactions required. Interactive demonstrations help expedite the agent’s learning process by providing a shared environment between the agent and the demonstrator to take turns in. This helps the agent learn directly from the demonstrator and allows the demonstrator to correct deviations that the agent made from the task. This approach is more natural to implement than other similar imitation learning techniques and has shown to reach a better level of performance faster than an enhanced Deep Q-Learning Network (DQN).

Language

English

Share

COinS