With the breakthrough in neural network techniques and the success of AlphaGo, we at Rayark think that the application of deep reinforcement learning (DRL) can already yield enough marginal benefits, and its technical requirements also tractable. For instance, during the game development cycle, we often have to adjust NPC AI due to changes in game mechanics, or that the game balancing tests are usually very incomplete in the early iterations. We figured that DRL can actually not only simplify the tasks for programmers, but also let designers rapidly receive more complete pictures of the results from their designs, thus speeding up the whole iteration of the process. Based on this idea, we are trying to use DRL in the upcoming TCG-MOBA game Soul of Eden, letting AI find the winning strategies by itself, and has got preliminary results. We hope to share what we've learned during our implementation of the techniques, difficulties encountered and some promising prospects in future game development. 

Note: This session will be conducted in Mandarin.