Martin Riedmiller : Increasing Autonomy in Self-learning Neural Control Architectures

Jeudi 11 avril 2013 de 14h00 à 15h30

Auditorium IRCICA, parc scientifique de la Haute Borne à Villeneuve d’Ascq

√ Abstract :

Humans are very good in perceiving all kinds of high-dimensional sensory inputs, extracting the meaningful information and acting on that information to pursue their goals. Having this in mind, our vision is a learning system, that takes raw, potentially highdimensional sensory inputs (e.g. raw image data), extracts the relevant information, and learns to act by experiencing success or failure.  In this talk I will provide some first successful examples along this line of research. In particular, I will discuss neural network based architectures and algorithms that are the basic building blocks of our neural control architecture.

√ Bio :

Future computer programs will contain a growing part of ‘intelligent’ software modules that are not conventionally programmed but that are learned either from data provided by the user, or from data that the program autonomously collects during its use. Successful realisations of this concept can be found in several of our application areas: robotics (our robotic soccer team Brainstormers became worldchampion at RoboCup 2005, 2007, 2008 (simulation 2D) and worldchampion at RoboCup 2006 and 2007 (Middle Size), and won the Technical Challenge Award at RoboCup 2006, 2007 and 2008), sales rate prediction or control of technical processes.

Les commentaires sont fermés.