Laboratoire Matière et Systèmes Complexes

Short introduction to 3 independent projects on applications of information theory, complex network theory, and machine learning methods

Detlef Holstein

Abstract: In the talk 3 topics of my former research work are shortly discussed: * Optimal Markov approximations and generalized embeddings:   A criterion for optimal generalized Markov approximations for memory in   effective dynamics of complex systems from information-theoretical entropies,   derived quantities and their statistical errors is introduced and justified. * Application of the theory of complex networks to neuroscience:   The role of network topology and noise on dynamic behaviour in neuronal networks   is analyzed. Phase transitions, a hysteresis phenomenon and synchronization of   neuronal activity are observed in our modeling approach. * Usage of machine learning methods in the framework of functional magnetic resonance:   The hemodynamic response function is predicted from the neuronal Calcium signal   with methods from the domain of artificial intelligence. The predictive performances   of machine learning methods from Echo state networks, multilayer perceptrons, and   shallow convolutional networks in the deep learning framework, and of the double   gamma function canonical model are compared.

Thursday January 23 at 13:30 in room 646A