Software

We are pleased to announce the availability of the first
sparse grid classification method.

This novel algorithm developed by PRUDENTIAL SYSTEMS (prudsys)
in cooperation with the University of Bonn is the first
non-linear classification and regression method
that scales linearly with respect to the number of
records and, thus, can be applied to millions of records.

It is based on a discretization technique calles Sparse Grids
which allows a discretization of smooth functions in higher
dimensions (up to 20). Sparse grids where developed in the 90ies
and are used for solving high-dimensional partial differential
equations. High-dimensional wavelet functions on sparse grids
are applied to solve regularization problems like that of support
vector machines but much faster and in a hierarchical basis
making the results interpretable.

The method is implemented in the prudsys DISCOVERER 2000, a universal
data mining tool developed by prudsys that will be presented
at the KDD 2001.

In addition, in the talk "Data mining with sparse grids using
simplicial basis functions" by Jochen Garcke of the University of Bonn
at the KDD (Research Track 5: Classification & Regression) the
sparse grid classification method is described.

Dr. Michael Thess                        phone:         +49-371-5347-580
PRUDENTIAL SYSTEMS SOFTWARE GmbH           fax:         +49-371-5347-126
Annaberger Str. 240
D-09125 Chemnitz e-mail: thess@prudsys.com
GERMANY WWW: http://www.prudsys.com/