The Bernstein Center Freiburg




Informal Seminar
Marcel Sauerbier

Faculty of Engineering
University of Freiburg


Cumulant Computation for Higher Order
Correlation Analysis in Neural Networks

Higher order correlation are computationally expensive when performed with k-statstics. A promising method to avoid a drop in performance is to express higher order correlations as theoretical cumulants of a point process data model instead. In this paradigm, however, all possible statistical correlations between single spiking events need to be taken into account. This is equivalent to enumerating all unique phylogenetic trees with a given number of leaves , representing those events. Said trees are then translated into multidimensional integrals, facilitating explicit cumulant calculations. An algorithm to compute these integrals is presented and its correctness proved.

Monday, March 24, 2014

11:00 h
BCF Library
First Floor
Hansastr. 9a
The talk is open to the public. Guests are cordially invited!
www.bcf.uni-freiburg.de