~ Nonlinear Adaptive Filtering as a Form of Artificial Intelligence ~
 Home History References Contact

## Reusable code

Kolmogorov-Arnold reusable code.
I decided that I should provide sort of reusable code for those who wish to test accuracy of Kolmogorov-Arnold superposition model. The test model is multivariate function $$z = F(x_1, x_2, x_3, x_4 , x_5),$$
where $F$ is combined from another five nonlinear functions: $$F = F_3[F_1(x_1) * F_2(x_2)] - F_5[F_3(x_3) * F_4(x_4)] * [F_4(x_5) - F_1(x_5)],$$ $$F_1 = 1.0 - exp(x) / exp(1.0),$$ $$F_2 = sin(3.14 x),$$ $$F_3 = 0.25 / (x + 0.2),$$ $$F_4 = |x - 0.5|,$$ $$F_5 = ln(x^2 + 1.0),$$ and arguments are random numbers from interval [0, 1.0].

Single Urysohn operator gives discrepancy near 5%, Kolmogorov-Arnold gives exact model with the error of a fraction of percent.