Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

BCCN/BFNT AG-Seminar

Tuesday, 21.06.2011 17 c.t.

Recognizing sequences of sequences using nonlinear dynamical systems

by Dr. Stefan Kiebel
from Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig

Contact person: Christoph Kolodziejski

Location

Ludwig-Prandtl Hörsaal, Am Faßberg 11, AI-Gebäude

Abstract

The free-energy principle is a general, mathematical theory for brain function. In my talk, I will describe mathematical models under the free-energy principle that lead to the hypothesis that cortex forms a hierarchy of time-scales along a caudal-rostral gradient. This cortical hierarchy recognizes sensory dynamics in the environment which were generated by coupled nonlinear dynamical systems at multiple time-scales. The empirical evidence supporting this hypothesis is briefly discussed. In addition, I will talk about another application of the free-energy principle, where we perform dynamical, online Bayesian inference based on standard recurrent neural networks. The resulting dynamics can themselves be interpreted as network equations. Using the example of recognizing human kinematics, I illustrate the potential usefulness of this technique (‘recognizing recurrent neural network’) for better understanding neuronal function and as the basis for novel machine learning approaches to state estimation with nonlinear, multidimensional dynamics.

back to overview