SPA Conference session: Asteroids: The machine war | |||
| One-line description: | Learn the fundamentals of Machine Learning by programming a squadron of fighters in Asteroids | ||
| Session format: | Long Tutorial [read about the different session types] | ||
| Abstract: | Asteroids is an arcade game in which you control a fighter that is bombarded by asteroids and attacked by flying saucers. It is a hostile environment that is ideally suited for Machine Learning. In this session we will provide a framework that will enable you to write the Artificial Intelligence of a (fleet of) fighters. A participant is provided with the means to extract sensory data from the asteroid fighter to provide as input to several machine learning algorithms, like neural networks and reinforcement learning, that will determine the behaviour of the fighter. It is your objective to score as many points as possible and outscore the other teams. Starting with limited possibilities in a controlled environment, a team's AI must complete objectives of increasing complexity to unlock the full potential of the Asteroids fighter capabilities. After completing the objectives, the full API is available and the written AI is transferred to the Asteroids arena in which other teams can be encountered. The constructed AI can now be expanded on, to create a squad of fighters that will roam the arena in search of points that can be earned by gunning down hostile objects. The controlled environment is still available and can be used to test new iterations of the AI, as well as running cycles to train the AI if the chosen algorithm requires it (e.g. Reinforced learning algorithms or neural networks). | ||
| Audience background: | Background in Software development and a inquisitive mind. Basic understanding of the language JavaScript. | ||
| Benefits of participating: | The participants will learn the fundamentals of Machine learning in a hands-on, engaging and competing fashion. | ||
| Materials provided: | We will provide a Virtual Machine that enables participants to compete in an open source Asteroids machine learning framework. | ||
| Process: | Participants are invited to pair up. Each team will receive a virtual machine with an interface to the Asteroids world. The teams will get to know the environment by following training session in which they need to program their fighter to achieve various goals. After the tutorial their program enters the Asteroids arena to compete with other participants AI. | ||
| Detailed timetable: | 00:00 - 00:15 Introduction 00:15 - 00:45 Setting up framework 00:45 - 02:15 Tutorial phase 02:15 - 04:00 Start of competion 04:00 - 04:30 Finale & Retrospective | ||
| Outputs: | Handouts explaining the theory behind the various machine learning algorithms Understanding of Machine learning Playback of the competition Sources to setup the game | ||
| History: | This session will have various try-outs before SPA 2014 | ||
| Presenters | |||
| 1. Daan van Berkel Luminis |
2. RobinVan Kaathoven | 3. | |