Hey guys! If you've been following me on LinkedIn you might have noticed that I recently finished a course on Reinforcement Learning (RL).

RL is a paradigm of Deep Learning that enables 'agents' to learn by themselves in an 'environment'. For example, take the open-world game of Spider-Man. The agent here is Spider-Man while the environment is the virtual world he's in.

At the core, with the approach of RL, we can train Spider-Man to swing around New York City all by himself. Ever wondered how YouTubers make AI play Super Mario Bros and Atari games? This is exactly how.

What excites me most is that OpenAI, the creator of ChatGPT and DALL-E 2 which are the two revolutionary AI services offered by them, offers a framework for RL.

It's called the OpenAI Gym, with some intuition you understand why it's called a 'gym'. The Gym provides pre-built environments which come paired up with an agent so you can use it out of the box and quickly gain traction with your experiments.

In my case, I used a similar framework offered by a collaboration between Unity and Hugging Face. A little about each: Unity is a game engine that game developers use to design and create interactive games. Hugging Face is a company that provides tools and pre-trained AI models to developers, contributing to the open-source community.

Both these tools complement each other to provide a brilliant framework that lets you train agents to interact with the environment. There are, of course, several algorithms available to use for this (policy-based, value-based, hybrid, etc.)

Thomas Simonini, a pioneer and a core contributor to the Reinforcement Learning field at Hugging Face. I took his course which gave me an in-depth overview of Reinforcement Learning.

The essence of Reinforcement Learning can be used for various applications, even for self-driving cars. The AI will collect RGBD (Red, Green, Blue and Depth) and LiDAR data (Point-cloud data) to learn from its environment. I cannot speak for the various advanced AI that is present in Tesla and other automobile companies but it fundamentally works on these principles.

This also opens up possibilities for the gaming industry. Imagine a dynamic boss battle that learns from the user's actions and behaves accordingly to make the game challenging and extremely interactive.

There is a lot to cover in Reinforcement Learning and I can't do justice to it writing about all of it in one post. I'll try to go more in-depth in my next post.