Deep Learning with Retro Video Games
Video games are one of the most exciting environments for the future of machine learning. Specifically, reinforcement learning models, that learn through interacting with their environment, have found a true training and testing ground in the world of video games. This is because reinforcement learning, unlike other areas of machine learning, is focused solely on a reward signal and maximizing a specific reward function. So we will first look at some of the recent accomplishments and talk about deep reinforcement learning at the highest levels of engineering like AlphaStar and OpenAi Five. Then we will get hands-on and look at creating our own deep reinforcement learning agents using OpenAI's Gym-Retro environment. This talk will show you how to setup your learning environment, introduce you to using convolutional neural networks inputs for your model (to read the screen), and show you how to build your first deep reinforcement learning model that learns to play a level in Sonic the Hedgehog 2 (and other retro video games).