Reinforcement Learning in the MiniHack Learning Environment

Reinforcement Learning in the MiniHack Learning Environment

Presenter(s)

Ian M. Cannon

Comments

Presentation: 1:20 p.m.-1:40 p.m., Kennedy Union 211

Description

Reinforcement Learning is a branch of machine learning in which a computer agent receives rewards for interacting with its environment by being given observations, formulating actions, and taking those actions in the environment. Reinforcement Learning has been used to exceed human performance in game-like environments from Checkers to Go to DoTA2. Many of these achievements were hard-fought by overcoming challenges in each environment individually. NetHack is a challenging game that makes an excellent Reinforcement Learning environment for its combinations of broad action and observation space with sparse rewards and general difficulty. In this work, we investigate reinforcement learning methods in MiniHack which is a mini version of NetHack. We will introduce challenges of the difficult game of NetHack and explain how MiniHack can break up this large environment into smaller, composeable, more tractable mini-environments. We have used this environment to solve challenging problems in Reinforcement Learning and will talk through these challenges and what we have done to overcome them.

Publication Date

4-20-2022

Project Designation

Graduate Research

Primary Advisor

Van Tam Nguyen

Primary Advisor's Department

Computer Science

Keywords

Stander Symposium project, College of Arts and Sciences

United Nations Sustainable Development Goals

Good Health and Well-Being; Quality Education

Reinforcement Learning in the MiniHack Learning Environment

Share

COinS