Published on 04.06.2019

Playing Atari with Six Neurons

The Best Paper Award at the AAMAS 2019 conference in Montreal has been awarded to a team from the Informatics Department of the University of Fribourg.

In computer science, conference presentations are often more prestigious than journal publications, and AAMAS (Autonomous Agents and Multiagents Systems) is a major international conference. The lead author Dr. Giuseppe Cuccu presented a paper addressing the complex problem of autonomously learning to play videogames. It focused on a simulator of the Atari gaming console from the early 1980s, which is often used as a benchmark for Artificial Intelligence programs.

Typically systems that learn to play Atari games are based on neural networks composed of hundreds of artificial neurons. The neural network presented by the Unifr team consists of only six neurons!  It is capable of playing a selection of different games by using clever tricks in processing the video images.

This work opens up interesting ways to make Artificial Intelligence faster and more efficient, and convinced the conference chairs in Montréal to award this work as the conference best paper.

Links :

G Cuccu, J. Togelius, P. Cudré-Mauroux: playing Atari with Six Neurons

AAMAS 2019 - International Conference on Autonomous Agents and Multiagent Systems