- Title: Analyzing State Spaces of Atari Games
- Speaker: Mark Nelson (American University)
- Time and date: 1pm to 2pm, June 28th, 2023 (Wednesday)
- Room: Virtual (Zoom)
The Game AI Research Group is glad to announce a (virtual) talk by Mark Nelson on Wednesday June 28th at 13:00. Members of DERI can also join us in PB.7, although there is limited space.
The Arcade Learning Environment (ALE) has been a popular domain for AI research over the past decade. It extracts current score and game-over status from an Atari emulator, making it easy to test search/planning and reinforcement learning algorithms on 104 classic arcade games. After a flurry of activity, it is now considered mostly “solved” as a state-of-the-art challenge, at least for deep reinforcement learning. But what exactly has been solved, and how well? And does this give us any generalizable information about algorithmic decision making? This talk will give an overview of some of my current work on better understanding what precise challenges are posed by ALE as a decision-making benchmark. My main strategy is to quantify aspects of each game’s state space, such as branching factor, branching structure, and frequency of meaningful decisions.
Mark Nelson is an Assistant Professor of Computer Science at American University in Washington, DC. His research focuses on AI and games from both technical and design perspectives. This includes using games as testbeds for studying sequential decision-making under time constraints on the one hand, and as creative design domains for automated and AI-assisted design on the other hand.