Billnet The Fast and Free Neural Network Simulator

BILLNET

The Fast and Free (GPL) Neural Network Simulator.

Last Update: 22 Jan 2005

What is Billnet?

Billnet is a modular neural network simulator platform. Its primary goal is to simulate as many as possible of the existing architectures and algorithms for neural networks in an effective and light weight manner so Billnet could be used simultaneously for research, educational and industrial solutions.

Where can I find Billnet?

Billnet is now hosted in savannah. You can download the latest CVS version from there. Older versions are currently not available.

Why to use Billnet?

Due to its modular nature we find Billnet very good for algorithm developing. Once the core platform remains the same it is very easy to compare different algorithms under the same conditions, and reliably decide which is better suited for the specified task. Billnet is also good for students who may want to take a closer look at the neural network technology or maybe implement an algorithm for training purposes. Billnet is small and fast, thus very effective for real world applications. It can be easily be part of an embedded or real time system. Finally Billnet has a proven track reckord.

Billnet's publications

Here is a list of all the publications Billnet's has been used and we are aware of.

What's new in the Savannah version?

convert to autotools; Now the famous ./configure; make; make install is possible. Partially tested with cygnus, Microsoft's msvc5 and Borland's compiler.
separate preprocessors, algorithms, and the Billnet core library are now dlls (on linux at least). That means:

an algorithm can call an another algorithm e.g LVQ (planned) calls to knn for its initialization
it will be possible for algorithms, preprocessors or custom programs to be developed outside Billnet sources.

codemap and BDF separation. This one eliminates nasty hacks from all over the Billnet source and from user's side scripts and phony data sets also.
new scanner/parser for configuration files. The concept of scenario is introduced in order to cover many use cases with one configuration file.
linear weight memory model for even greater efficiency
threshold, weight vector unification
a general clean up

TODO

Algorithms in CVS version	Algorithms in version 0.43
perceptron regression adaline	perceptron regression adaline
back_prop with CrossEntropy Cost Function with Sigmoid Output Function with Tanh Output Fucntion	back_prop
aleco aleco2	aleco aleco2
rprop	rprop
knn fast knn by Nene-Nayar	knn fast knn by Nene-Nayar
conj_grad*	conj_grad
	dbd (delta bar delta)
	quickprop
bfgs*	bfgs
	adaptive
dogleg*	dogleg
lev_mar (Levenberg-Marquardt)	lev_mar (Levenberg-Marquardt)
FLF3	FLF3
Preprocessors in current version	Preprocessors in version 0.43
shuffle	shuffle (built in)
part	part (built in)
normalizer	normalizer
	randomizer
	addnoise
	hiorder
	app2class
	salience
	kmeans (as algorithm)
	SOM (as algorithm)

* WARNING: There are some discrepancies, the alogrithm is either not stable or fully debugged yet

Planned

RBF
LVQ
LMAM
OLMAM
SVM
FLF split up to library parts (Done)

More work

Loads of Documentation

LICENSE

Billnet is free software with the GPL meaning of the term. Read to learn what is GPL.