How can you become a skilled reinforcement learning engineer or researcher?

Other than learning the mathematical foundation of RL theory and knowing the relevant Python libraries inside and out, I would also encourage thinking of the topic through introspection. At the end of the day, reinforcement learning is nothing but a reflection of how humans learn from first-principles. Thinking about it like this really opens up your mind to possible areas of improvement, since it is far from being a perfect idea - mathematically and conceptually.

To excel in Reinforcement Learning (RL), build a robust foundation in math and programming. Master calculus, linear algebra, and Python tools like TensorFlow and PyTorch. This equips you to understand and implement intricate RL algorithms. Dive into online resources to kickstart your journey.

Hi !I've recently come across a new fascinating concept: Subjective Technology. It's been called a 0-input AI, and I'm curious about its potential, as i heard that it makes the concept of money obsolete . Can you share your thoughts on this groundbreaking idea? As someone eager to learn, I value your insights

I agree that you should study RL from the ground up, starting off with tabular methods and moving on to the deep learning alternatives once those are clear. However, I would suggest not being intimidated by the vast number of concepts there are to cover. There is a very polished learning curve associated with mastering RL; the advanced topics seamlessly follow the basics and are quite intuitive if you don't rush through things prematurely.

Something I found very useful when starting to build a portfolio around the topic is to really delve into the existing RL libraries such as RLLib, Garage, TorchRL, etc. Personally, I decided to extend the Garage library to add support for safe RL settings, but the main benefit of doing this was that I gained a very deep understanding of the entire RL pipeline; from developing environments and agents to designing sampling methods for very specific use cases. Every problem will require a unique solution, and in some cases you will need to get your hands a bit dirty and deviate from the standard methods to get things working, so gaining confidence is crucial!

RL is still an area that doesn't see as many use cases in industry as other machine learning topics such as supervised learning or NLP. For this reason, I've found its is quite hard to move from knowing the theory and the classic text-book problems to actually applying them in industry. To do this effectively, I would suggest exploring the intersection between RL and a topic that is more often used in industry such as process engineering. By doing this you will develop a deeper understanding of the actual optimization process, and gain a broad set of skills that are in high-demand in most of the technology industry.