SMU Guildhall Graduate Catalog Spring 2018 — Cohort 26 - Page 96

Tianyi Zhao Software Development « Reinforcement Learning Agent for StarCraft II Mini-Games My thesis aimed to build reinforcement spent six months on this project, which games. When I built the agents in my and reinforcement learning algorithms. learning agents for StarCraft II mini- project, I used two different reinforcement learning algorithms. One of the algorithms An agent which can learn and practice (A3C), which is state-of-the-art, and the Strategy games. It can help developers is Asynchronous Advantage Actor-Critic other is Deep Q-Learning. Both agents can learn and play all StarCraft II mini- games without changing any parameter. I chose this topic because I am very passionate about machine learning in gaming and I am a big fan of StarCraft. I 96 gave me an opportunity to learn Python SOFTWARE DEVELOPMENT with itself is a great tool for Real-Time balance the game and explore more new strategies.