Most Real-world problems contain uncertain and noisy data. Converting data of this type to symbolic
predicates removes most of the noise, but also lots of knowledge. Soft computing uses numeric
data in a more direct way. As a consequence, such methods can solve problems which would be
unsolvable by traditional algorithms.
In the last fourty years, more and more methods of this type have evolved, descending from the
fast-growing Data-Mining and Machine-Learning sectors.
As the name says, this is also the core-business of Subsymbolics.
Our know-how includes:
Time-series processing, classification and extrapolation
Morpholigic and Polynomial adaptive filters
Neural Networks (Feed-forward and Recurrent)
Classification methods based on Eigenvalues
Support Vector Machines
Since most numeric algorithms depend on fast floating-point math, our solutions are based on standard libraries like
MTL, GMP, BLAS, LaPack, PVM and MPI to exploit the capacity of the different types of hardware, starting from small x86
systems up to high-end Vector-Processors and Cluster-Computers.