Over the past few years, reinforcement learning (RL) has achieved promising results and it is currently being explored in a wide range of fields. In areas such as self driving cars, gaming and medicine, RL is the frontier of state-of-the-art results. In this workshop we will explore what the fuss is all about!

This workshop serves as an introduction to reinforcement learning where the participants will implement a Pac-Man agent. The Pac-Man agent will learn how to solve different maps using Q-learning and Deep Q-learning. We start out by exploring Q-learning, a cornerstone in RL. Expanding further, we continue on to deep Q-learning, which utilizes neural networks. The code is executed in the cloud on Jupyter notebooks, and for training the neural networks we use GPUs in the cloud. Everything is written in Python.

No prior knowledge of reinforcement learning is necessary.

If reinforcement learning has been a mysterious domain to you, this session will most likely leave you with a greater understanding of the process and aid you in how to set up projects of your own.

Optional preparation for the tutorial
The tutorial will be much like a walkthrough, so it is quite fine to just follow along without programming yourself. If you do, however, want to interact with the actual code, it is recommended that you clone the project and set up the Python environment beforehand. While most of the tutorial will be in Jupyter Notebook, some setup is also required on local machines. Since we only have 1.5 hours, we will not have time to help individual participants setting up the project during the actual tutorial, but if you do face issues you can send us an email ahead of the tutorial. We will be happy to help!

Project repository: https://github.com/knowit/ml-pacman
Manu: manu.gopinathan@knowit.no
Malte: malte.loller-andersen@knowit.no