Machine Learning Arena: Creating an ML Based Game
Smith, Gillian Margaret
Whitehill, Jacob Richard
We present a new game, “MLA: Machine Learning Arena”, in which the player’s goal is to train a machine learning agent to win a boxing match. The game features multiple phases, fully animated characters for the player to control, and machine learning integration. We tackled several technical and design challenges, including: 1) Communicating machine learning progress through UI elements to the user, 2) Training an effective model for the game agent despite poor training examples from users, 3) Explaining key ideas about machine learning to players with no background in the field. We conduct user testing with 27 players to determine if they feel that the ML is learning from them. We found that 85 percent of players were able to distinguish between a random agent and the trained agent.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.