YANNS v 1.0

Yet Another Neural Network Simulator

Romuald Boné (romuald.bone(a)univ-tours.fr) and I (michel.crucianu(a)inria.fr) are happy to announce the availability of (yet another ;-) neural network simulation tool presenting the following general characteristics:

Yanns is using a simple description language;
Yanns is written in multi-platform C++ and the code is periodically debugged with the help of professional tools;
while directed towards the simulation of recurrent neural networks, Yanns lets you build very diverse network architectures (with or without layers, with or without recurrent connections, simple, FIR or IIR connections, etc.);
the best-known algorithms for supervized learning (in feedforward or recurrent networks) are implemented in Yanns, together with some more recent algorithms; the object-oriented development allows these algorithms to work on various network architectures.

More information regarding the conception of Yanns can be found in the following papers: NEURAP'98, ICONIP'98. Details concerning the description language, file formats, etc. can be found in the User manual in HTML, in PDF or compressed PDF. Unfortunately, the manual is only available in French, but the description language is using English words. Also, a compiled version is directly available for a linux platform and for a wintel platform.

A simple example (the classification of iris...) is available in the files iris.dsc (description file), irislearn.set (learning data), irisstop.set (early stopping data) and iristest.set (test data). To try it:

system>yanns iris.dsc > result_file.txt

We can send you the sources (the comments are mostly in English; please address yourself directly to romuald.bone(a)univ-tours.fr), as well as the programming manual (in French), but we may be unable to engage in maintaining a multi-source software. Also, we may be unable to provide very prompt answers to your questions.