My main job is to design game AI for a real-time strategy game Soul of Eden.
More specifically speaking, I designed a deep reinforcement learning (RL) model based on Asynchronous Advantage Actor-Critic (A3C) algorithm without any supervised data. My model outperformed all of the rule-based models previously used by the company, with win rate more than 80%. I not only built the model but integrated the model (written in Python) into the real game (written in C#).
Formosa Grand Challenge is the largest national AI competition in Taiwan, funded NT$100,000,000 (US$3,500,000) by the Taiwan government. We successfully attracted 100+ teams to participate in the warm-up competition.
My main job was to provide baseline for the competition and to design the questions for the competition. Specifically, I designed a seq2seq with attention model for the baseline, processed raw data collected from real world TV programs in Taiwan and then designed the questions.