Im Rahmen des Forschungsseminars laden wir recht herzlich ein zum Vortrag:

Dr. Ronald Ortner (Montanuniversität Leoben)
"Learning based on Regrets"

Abstract
In reinforcement learning, a learner acts in an unknown environment and has the task to learn optimal behavior by evaluating feedback of the environment. This talk will first present various reinforcement learning settings and introduce learning algorithms for which performance guarantees can be given. A particular focus will be put on the so-called regret as a performance measure for learning algorithms. The regret quantifies how much a learning algorithm loses with respect to an optimal policy that knows the environment in advance. In the second part of the talk, the problem of model selection in the context of reinforcement learning will be considered. It will be shown that learning algorithms can successfully learn without trying to identify an underlying true model.

Speaker

Ronald Ortner received a Master in philosophy and a PhD in mathematics from the University of Salzburg. After working two years in the semiconductor industry, he changed back to academia and took a position as assistant professor at Graz University of Technology in 2002. In 2003 he became assistant professor at the Chair of Information Technology at the Montanuniversität Leoben. Since 2010, he is associate professor at the same institute. He has more than 30 refereed publications in scienti c journals and conferences, mainly in the area of reinforcement learning.