So you’re an absolute beginner who wants to learn about Reinforcement Learning.
Let’s start with what reinforcement learning, or RL, is. RL is a process where an agent interacts with an environment and gets rewards. The agent then uses those rewards to slowly get better at the task that’s being rewarded. It’s a simple concept, but the process can result in superhuman agents, like AlphaZero and AlphaGo, that beat the world’s best human players in games like Chess and Go.
If you’re here though, you probably already know that and you’re trying to figure out where to start. What I’m going to try to do is list the courses I recommend, in order, to get into RL as quickly as possible. All of these courses can be found on Udemy, usually on sale for under $20.
It’s worth remembering that RL is an advanced topic, and the road is long but rewarding. I’ll also note that I’m an AI enthusiast, not a professional. With that out of the way –
Python
Python is, without a doubt, the most popular programming language for AI projects. If you’ve never programmed before, I recommend starting with Tim Buchalka’s Learn Python Programming Masterclass. It’s a ground up guide that will teach you the basics and give you enough practice to get really comfortable with the language. At this stage, I’d advise against using tools like ChatGPT to learn to code. You’ll get good results without the muscle memory you’re really trying to build.
https://www.udemy.com/course/python-the-complete-python-developer-course
Machine Learning
If you already know Python, or you’re comfortable learning on the go, it’s time to jump into ML. Machine Learning is the art of training a machine to make decisions based on past data instead of explicitly coding in rules. My favorite recommendation here is Mike Cohen’s A deep understanding of deep learning. He’s going to build up ML concepts piece by piece and help you gain an intuitive understanding of how to work with models. This is a longer course, at 60 hours, but this is another place where time spent is actually a win. Spending lots of time building, tuning, and working with models to solve problems is going to build up a skill set that will serve you well later in the process.
https://www.udemy.com/course/deeplearning_x/?couponCode=24T4MT180225
Math
Alright, I tricked you a little bit because there was some math in Mike Cohen’s course. I actually recommend waiting for the middle of the process to dive into the math, because starting there can lead to discouragement when you don’t understand why you’re doing it. Machine Learning makes use of Calculus, Linear Algebra, and Statistics. Out of the three, Linear Algebra and Statistics are the most important in understanding what’s going on with your model. Calculus is necessary if you want to understand the back propagation more deeply, but I would consider that an advanced topic.
If you’ve gotten this far and want to brush up on your math, check out Krista King’s Linear Algebra and Statistics courses.
https://www.udemy.com/course/linear-algebra-course
https://www.udemy.com/course/statistics-probability
Reinforcement Learning
Now for the fun stuff! Reinforcement learning. I have a hard time making a single recommendation here, but what I’m going to suggest is Phil Tabor’s Modern Reinforcement Learning: Deep Q Agents (PyTorch & TF2). Phil takes a “papers to code” approach that teaches you to start from a research paper and build out the solution. I’ll admit that I find his courses a bit dense, but as you progress, you’ll be learning from someone working on the cutting edge.
https://www.udemy.com/course/deep-q-learning-from-paper-to-code
Build Things
One of the best ways to learn is to just go build things. Start with someone else’s solution, and then figure out how to change it and solve a different problem, and this is where I’ll plug…me! I’m a fan of AI, and I post RL & Robotics projects on YouTube. Check out some of my videos at the link below –
https://www.youtube.com/@robertcowher