Agenda aka Next few Learning Nuggets 1

Let me start out by expressing my overall intent in writing this — this just creates a forcing function to push me to dig deep and learn more of Reinforcement Learning. The way i am going to do that is to keep creating a list of future articles that i intend to write :-)

Here goes my first list:

1. Trial-and-error search and Delayed Reward are supposedly the two most important distinguishing features of Reinforcement Learning. Let’s dig deep on each of these in separate articles. For example: how does trial and error work with another challenge in RL — Explore and Exploit ? Delayed Reward — i would love to understand how the quantum of delay would change how we solve different problems ?
2. Reward function vs Value function. The first generally indicates what is good in an immediate manner while the latter is the sum total of rewards accumulated over the future, starting from that state. Rewards are generally given directly by the environment, while the Value of states are estimated to solve the Reinforcement Learning problem. I am interested in learning more about the relationship between these two functions in a few case studies.
3. Evolutionary methods do not fall under the set of methods that are covered under Reinforcement Learning. How do they differ ? Are there problems that they do a much better job than Reinforcement Learning methods ?
4. A model of the environment can help immensely in deciding a course of action. What is Model-free ? When do we go for Model-free methods deliberately and where are we forced to do that ?