Over the course of the previous five chapters, we learned how to evaluate the value of state and actions for a given finite MDP. We learned how to solve various finite MDP problems using methods from MC, DP, Q-learning, and SARSA. Then we explored infinite MDP or continuous observation/action space problems, and we discovered this class of problems introduced computational limits that can only be overcome by introducing other methods, and this is where DL comes in.
DL is so popular and accessible now that we have decided to cover only a very broad overview of the topic in this book. Anyone serious about building DRL agents should look at studying DL further on their own.
For many, DL is about image classification, speech recognition, or that new cool thing called a generative adversarial network (GAN). Now, these are all great applications of DL, but, fundamentally...
