Res Neuronum - Features

Res Neuronum - Simulation of Neural Networks

News

Features

Manuals

Translation

Download

Contact

About

Features
Res Neuronum is a software simulating different kinds of Artificial Neural Networks which can be included as modules. In the same way it is possible to write modules not only demonstrating how various parameters have to be set up and how the networks are functioning but rather showing better their usefulness within practical applications such as speech recognition. An example is included. In version 1.02 the only network implemented is a Perceptron (MLP) with unlimited number of layers/neurons (i.e. limited only by range of variables) and the Backpropagation algorithm as learning rule. Within this dynamic link library "BackPropagation.dll" you'll find features to construct a network and a PatternSet consisting of many Patterns, to train the network (learning of weights, slope and thresh; momentum; recursive learning-rate and adaptive learning possible), evaluate the learning data (weights, activity etc.) and finally recall the whole net for Patterns of your wish. As an extra bonus an optimization-algorithm was implemented; it was developed and extended at the TU Ilmenau. Moreover a second DLL "SpeechRecognition.dll" was created; it demonstrates how BackPropagation Networks can be used to learn Speech, i.e. single spoken words, even without the help of other Networks (such as Kohonen Feature Maps). For further details please read the manuals (which are in German only unfortunately). Details are given in the thesis available over the library of the University of Ilmenau (TU). Its title is "Entwicklung eines Praktikumsversuches zu Neuronalen Netzen auf der Basis von Borland Delphi".

Features

Res Neuronum is a software simulating different kinds of Artificial Neural Networks which can be included as modules.
In the same way it is possible to write modules not only demonstrating how various parameters have to be set up and how the networks are functioning but rather showing better their usefulness within practical applications such as speech recognition. An example is included.
In version 1.02 the only network implemented is a Perceptron (MLP) with unlimited number of layers/neurons (i.e. limited only by range of variables) and the Backpropagation algorithm as learning rule.
Within this dynamic link library "BackPropagation.dll" you'll find features to construct a network and a PatternSet consisting of many Patterns, to train the network (learning of weights, slope and thresh; momentum; recursive learning-rate and adaptive learning possible), evaluate the learning data (weights, activity etc.) and finally recall the whole net for Patterns of your wish.
As an extra bonus an optimization-algorithm was implemented; it was developed and extended at the TU Ilmenau.
Moreover a second DLL "SpeechRecognition.dll" was created; it demonstrates how BackPropagation Networks can be used to learn Speech, i.e. single spoken words, even without the help of other Networks (such as Kohonen Feature Maps).
For further details please read the manuals (which are in German only unfortunately).

Details are given in the thesis available over the library of the University of Ilmenau (TU). Its title is "Entwicklung eines Praktikumsversuches zu Neuronalen Netzen auf der Basis von Borland Delphi".