Founder and Chief Scientist of Blue Yonder
Prof. Feindt’s main experience and most successful research work is data driven software development, understanding and learning effects through data analysis, usage, further development and discovery of new multivariate statistical algorithms and respective software (today labelled “predictive analytics” or “machine learning’’) in big data environments. Real big data in modern particle collider experiments consist of PetaBytes of data per second and have to be reduced by a factor of about 1 million by automatic trigger systems before even stored to disk. Reconstruction and Monte Carlo simulations also run on the world wide Grid, predecessor of today’s commercial Cloud Computing.
Important work lies in the abstraction from the particular problem for an easy and successful transfer of these methods to other disciplines, and big data challenges in retail, various industries and financial services as well as real-time targeting mechanisms i.e. in online ad business.
In 2000 Prof. Feindt invented the NeuroBayes®-algorithm to predict future events by learning from samples of past events and developed it for usage in research and very different industries and applications like insurance, trading, retail, wholesale, finance, medicine, industry, customer relations. Key features of NeuroBayes® are its general approach for retail, industry and financial service problems, its robustness, speed, scalability and flexibility.
It cannot only predict the probability of binary events (final answer is yes or no, prediction output is probability of ”yes”), but also the complete probability density distribution of real-valued variables like turnover or price, which allows e.g. an optimal decision under uncertainty – a situation very common in almost any company.
At Blue Yonder he invented a number of further algorithms important for applications in retail and industry, now integral part of Blue Yonder’s proprietary machine learning library. These are e.g. the cyclic boosting algorithm for “explainable” predictions (vs. black box), an algorithm for isolating causal effects in historic data, and OR-by-AI using a combination of ideas from reinforcement learning, deep learning and NeuroBayes. Another highlight is the implementation of the NeuroBayes expert algorithm in FPGA hardware, in collaboration with KIT, for the Belle II experiment. Such a chip can calculate 8 billion intelligent decisions per second for deciding which parts of a sensor should be read out.
Prof. Feindt founded the company Phi-T in 2002 and Blue Yonder in 2008 to professionalize and further develop NeuroBayes® and attracted substantial seed investment capital. Phi-T GmbH was also managing a market neutral 100 M€ low-risk absolute-return investment fund completely automatically on the basis of NeuroBayes®-predictions until it was stopped due to legal changes. Prof. Feindt currently is on leave of absence from the Karlsruhe Institute of Technology KIT, continuing the research work but having no teaching obligations and works 100% as Chief Science Officer for Blue Yonder.
Research career at a glance
• Since 1997 Prof. Dr. Feindt is professor of physics at the University of Karlsruhe (Germany), now called Karlsruhe Institute of Technology (KIT). He supervised more than 60 PhD theses and 70 diploma/master theses, and co-supervised another 100 theses. He leads a research group of about 20 scientists at KIT. Currently he is on leave of absence from KIT for working full time for Blue Yonder.
• A recent highlight was
o the invention, implementation and application of a “full reconstruction factory” for the Belle-experiment, a software that performed roughly the equivalent of about 1100 Ph.D. theses fully automatically using about 70 NeuroBayes® networks and
o A new version just released in 2017 for the successor experiment Belle II, written by PhD students of him, leads to another doubling of efficiency with respect to human work, together corresponding to a € 2.1 billion gain in running cost.