Reinforcement learning for game personalization on edge devices

Good progress has been shown recently in the area of active learning, specifically, Reinforcement learning (RL). In this paper, the authors show how RL can be used to personalize games based on user-interaction with the game. The work uses Deep Q network models (DQN) and the open source framework OpenAI to build an RL model that is able to optimize the gamer's engagement level in a game. The authors define an example quantitative measure of gamer engagement and incorporate that into the DQN learning reward function. The gamer experience optimization is empirically demonstrated using a game of Pong. Simulation testing and analysis of results indicate adapted RL models increase engagement reward values, thus enhancing gamer experience. The contribution of this paper is twofold: (1) using RL, it paves the path for wider adaptation to user-behavior, starting with gaming, and (2) it shows analysis and feasibility of an RL algorithm on an edge device (Personal Computer) in real-time.