Efficient Exploration for Model-Based Reinforcement Learning

点击翻译按钮获取中文摘要

Reinforcement Learning (RL) is a branch of Artificial Intelligence (AI) research that focuses on developing algorithms which enable an agent to learn a certain task in the environment they are placed in, such as maximizing the score in a game. A crucial component of this learning process is for the agent to explore strategies that result in the best possible score. When these agents are not able to explore efficiently, their applicability becomes very limited for many use cases. The proposed study will focus on developing novel approaches that allow RL agents to efficiently perform exploration, using the most promising direction of the sub-field: model-based RL. On successful completion, this study will be of invaluable impact to the gaming industry, as it will allow feasible development of RL systems that can acquire high-level playing skills in many types of games, allowing them to be used as challenging opponents, and for extensive game testing.