Events in Physics
Prof. Michael Feindt (KIT Karlsruhe, Germany)
NeuroBayes - Predictive Analytics for High Energy Physics and "Real Life"
In this Colloquium I will talk about NeuroBayes, a robust, fast and award winning algorithm with exceptional generalization properties for classification and event-by-event unfolding. NeuroBayes has its roots in high energy physics. Since 2002 it is professionally further developed and applied in insurance, finance and trading industries by the private companies Phi-T and Blue Yonder run almost exclusively by high energy physicists.
Some successful applications from the high energy physics experiments DELPHI, CDF II and Belle are presented. Special emphasis is placed on a hierarchical full reconstruction ''factory'' that fully reconstructs 1042 decay channels employing 71 NeuroBayes networks in a sort of automatized meta-analysis, improving the effective statistics compared to the classical approach by more than a factor 2.
In addition, an overview over the wide diversity of industrial applications is given.