NeuroBayes – Big Data Predictive Analytics for High Energy Physics & "Real Life"

Michael Feindt, Karlsruhe Institute of Technology (KIT) & Blue Yonder

NeuroBayes is a robust, fast and award winning algorithm with exceptional generalization properties for classification and event-by-event unfolding (conditional probability densities) on the basis of big and smaller amounts of data. NeuroBayes has its roots in high energy physics and is used very successfully in many experiments at CERN, Fermilab and KEK, up to completely automatic hierarchical reconstruction and analysis systems consisting of about 70 networks that outperform classical and hand-made methods by more than a factor of two on thousand decay channels. Since 2002 NeuroBayes is further developed and applied by the private company Phi-T. Its daughter Blue Yonder, founded in 2008, has managed to close the gap between theory and practice and has a huge track record of extremely successful prediction and optimization projects in very different industries such as retail, online trading, insurance, finance, healthcare and industry, always completely data-driven using strictly scientific – mostly Bayesian – methods. Blue Yonder by now forms the largest industry pool of PhD data scientists in Europe, most of them with strong research experience in either physics or computing. The talk gives an overview of exciting projects in fundamental research and economy and demonstrates some of the difficulties in the interface between science and the "real world".