COIF-NN: Cloud Onset Instant Foreteller - Neural Network
One of the modules of the COIF-NN
GMS
(Generalized Multidimentional Splines)
GMS-module is
a novel smooth estimator capable of fast learning,
provided the (noisy) values of vector function are given along with
vectors of its gradient on a set of irregular points in a multidimentional
space.
A few examples are displayed below.
Here the one-dimensional function is dispayed:
Noiseless data (values,- marked with impulses,- and gradients) are given
Noisy data (values
are
given with 25% errors)
Here the two-dimensional function is dispayed:a) true function (given 3 points, marked with triangles),
b) approximation of the function, reconstructed by the use of 3 function values
c) reconstructed by the use of 3 function values and 3 gradients
Note: neither trigonometrical, nor polinomial functions are used in the strategy of the GMS.
