NEURO CODE






NEURO CODE can generate automatically C-code libraries ready to compile, for any operating system.

From any non-linear model or any neural network created by Neuro One (or within Neuro One formalism), Neuro Code generates automatically open source libraries in C-code, recompilable in any environment: Windows, Linux, large systems, microsystems and embedded systems as well as programmable robots. The software has been successfully compiled with several compilators on Windows/Pentium, Linux/Pentium, Linux/Itanium, HP11-Unix/Itanium and on AIDIG cards (cards for some industrial robot) and AIDIAG Premium (cards for Schneider Premium) produced by Prosyst. Tests are being done for embedded systems.

The libraries Neuro Code generates are:

• The transfer function of the non-linear model or the neural network;
• The confidence intervals calculated at the first-order for any value predicted by the transfer function;
• Tuned Levenberg-Marquardt algorithm for online training.

This library can consist of 1.500 to 3.000 lines of code depending on whether the studied model is static or dynamic. The Levenberg-Marquardt algorithm included in this library gives back the same results after learning as those given by Neuro One, with an average accuracy of eight decimal places. On the other hand, it is much faster, since it is optimised for the studied model. For identical models and training data sets, learning execution is ten times faster than with Neuro-One.

Neuro-Code is a neat solution to free oneself from Windows, or to work on both Windows and non-Windows operating systems. Neuro-Code is also interesting for repetitive work or to speed up huge databases processing. Finally, Neuro Code unify in one program all non-linear models treatments, including neural networks.


Please contact us if you are interested in Neuro Code and if you want a demonstration.

© Netral - June 2007