Deep Reinforcement Learning for Computer Games
Time: 12-16 August 2019
Duration and credits: 1 week, 3 ECTS (+ project work, 2 ECTS)
Teaching language: English
Level: Master and doctoral students
Max. number of attendees: not limited
Course coordinator: Ville Hautamäki, email@example.com
Responsible department: School of Computing
Learning outcomes: Basics of machine learning, autonomous agents (like robots).
Basics of machine learning (one day + practicals)
The basics of supervised and unsupervised learning. Supervised learning portion will first cover the task definition and classical approaches. But we will quickly move to modern deep learning -based models and algorithms. We will cover following models: classical feedforward, convolutive and recurrent networks. In addition we will cover generative adversarial networks (GANs) and variational autoencoders (VAE). This section will also include hands on portion.
Autonomous agents (two days lectures + practicals)
At first reinforcement learning will be introduced and the basic techniques will be reviewed. We wil then move to deep reinforcement learning (DRL), where RL can be easily applied to visual input tasks. The focus in the course is how to teach autonomous agents to play computer games, so we will show and use in practice some successful DRL learning environments, such as VizDoom, Unity OTC and so on. This section will also include hands on portion.
Autonomous agents project work (two days)
During this section students will develop (using Python) and train an autonomous agent that will be able to play a computer game. We will offer an learning environment for Toribash fighting game so that students will be able to easily start with developing their own agents. Agents then will be playing against each other and human players. Students will be offered a realtime LeaderBoard, where they can follow the progress of their agent (how well it plays against the other agents).
Modes of study: Lectures, practicals, learning diary
Study materials: delivered during the course
Evaluation criteria: pass/fail
Teachers: Lectures: Ville Hautamäki (UEF), Abraham Woubie (UEF), Anssi Kanervisto (UEF) and guest lecturers tba. Practicals: Ville Vestman, Anssi Kanervisto, Trung Ngo Trong, Ivan Kukanov