Press releases

08. February 2012

The traditional Swiss mail order company Ackermann uses prudsys real time recommendations
read more


07. February 2012

New features of prudsys RDE 3.1 Releases thanks to comprehensive parallelisation
read more


11. January 2012

DATA MINING CUP in Berlin: Federal capital set to be venue for world's largest data mining competition and prudsys User Days 2012
read more


Presse Archiv

x
Press releases

Press releases - Details

x
30. October 2003 09:25 Age: 8 Jahre

Prudsys receives US Patent of Sparse Grid Classification

By: Sandra Koegel

The prudsys AG has received the US patent for regression and classification methods based on Sparse Grid approximation. The Sparse Grid technology allows for the first time to apply non-linear classification methods, like Neural Networks and Support Vector Machines, to nearly unlimited data volumes.

The prudsys AG has received the US patent "Device and method for generating a classifier for automatically sorted objects" for Sparse Grid classification. This regression and classification method was developed in close cooperation with the working group "Scientific Computing and Numerical Simulation" of Professor Michael Griebel of the University of Bonn. The Sparse Grid technology is one of the most promising approaches in Data Mining because it allows the high-quality analysis of huge data volumes.

The Sparse Grid technique is an approximation method which allows an efficient discretization of high-dimensional functions. The theoretical fundament of the Sparse Grid approach was originally developed in the sixties by Soviet mathematicians; in the nineties the universities of Munich and Bonn had shaped the numerical basis of the Sparse Grid approximation and made the first implementations. The Sparse Grid technique uses wavelet bases which are constructed upon a hierarchy of anisotropic grids. This allows e.g. to solve 15-dimensional differential equations or to compress signals on multiple channels (like video).

The ability to solve high-dimensional operator equations efficiently, turns Sparse Grids into an ideal instrument for Data Mining applications. This idea has lead to the development of Sparse Grid classification which is realized in close cooperation between prudsys and the University of Bonn since the end of the nineties. Support is given by the WestLB bank, especially in the field of prognoses of financial data; and since summer of 2003 the German Cancer Research Center uses Sparse Grids in genomic research.

A first version of the Sparse Grid classification is implemented in the leading classification tool prudsys DISCOVERER and in the XELOPES Data Mining library. There are further patent applications of prudsys in the field of Data Mining and general Data Mining methods.



x
x x x