GRASP Special Seminar: Tim Genewein and Jordi Grau-Moya, Max Planck Institute for Intelligent Systems, "Decision-Making with Information Constraints: Free Energy Foundations and Applications"


A key characteristic of intelligent systems, both biological and artificial, is the ability to efficiently adapt behavior in order to interact with the environment for their benefit. Importantly, these systems are subject to information processing limitations due to both, their own computational constraints and the lack of knowledge about the environment. In this talk we give an introduction to a framework for optimal decision-making under such information constraints that is grounded in information-theory. The generality of the proposed approach allows modeling important aspects of the decision-making process based on the same first principles. In particular, the framework can describe not only bounded rationality i.e. decision-making with computational limitations, but also, model uncertainty (lack of knowledge about the environment), risk-sensitivity, hierarchies of abstractions and lastly, likelihood synthesis for perception action systems.

In the first part of the talk we will give an introduction to decision-making with information constraints whereas in the second part we will show an application to planning in unknown Markov Decision Processes and to hierarchical decision-making scenarios.

Presenter's biography

Tim Genewein received his BSc and MSc degrees in Telematics (information and computer engineering) from Graz University of Technology. He is currently working as a PhD student in the “Sensorimotor Learning and Decision-Making” group at the Max Planck Institute for Intelligent Systems in Tuebingen. He is particularly interested in hierarchical decision-making and the emergence of hierarchies of abstractions as a consequence of optimal acting under limited computational capacity.

Jordi Grau-Moya is a PhD candidate working on decision-making with information constraints at the Max Planck Institute for Intelligent Systems, Tuebingen (Germany). In particular, his investigations focus on decision-making with limited computational resources and in unknown environments, using statistical physics and information theory. Previously, he obtained his Industrial Engineering degree from the Polytechnic University of Catalonia in 2011 with final thesis on the topic of spiking neural networks.