User manual
The idea of artificial neural networks is based on attempt to simulate some properties and functions of brain. Therefore neural network can be viewed as simplified model of brain consisting of processing units shortly called neurons. Neuronix makes it possible to design, create and run such nets quickly. Knowledge necessary for problem solving is acquired at the stage of training. While training network weights are adjusted after training patterns presentation. Each single training pattern consists of target answers for pattern tha is input. The training is stopped when error rate is sufficiently low.
Potential fields of Neuronix application:
- Credit risk assessment,
- Forecasting of financial results and markets,
- Sales forecasting,
- Forecasting of stock prices,
- Pattern, handwriting recognition,
- Data analysis,
- Cptimization of some complex computational problems,
- Classification of objects,
- Acoustic signals analysis,
- Filtering the disturbance of signals
Main functions of Neuronix:
 supporting process of neural network design, starting from gathering data, then generating tarining and testing files; training and testing, until running the net using real input data
some steps of neural net creation are fully supported by special dialogs and utomatic procedures, so to make the work easy and user-friendly,
input values can be either numerical or textual (the last are converted automatically),
possibility of building hybrid applications using Neuronix and PC-Shell expert system,
Neuronix has the same built-in translator and algorithmic (procedural) language as PC-Shell,
dynamic visualization of neural net structure with input and output,
monitoring, visualization and automatic recording of training parameters,
Neuronix is equipped with special dialog window (creator) simplifying preparation of new project: :
 
In the case of using project creator the following files will be generated:
The possibility of using as input or output both the numerical and textual (symbolic) values is very useful feature of Neuronix while modelling. It gives the chance to communicating using linguistic expressions in more human-like way. Such models are often more realistic, especially if some factors are rather qualitative ones than numerical in nature.
To build model of any process user is not expected to know explicite the laws that rules behaviour of the process, e.g. in the form of expert knowledge or mathematical equations
It is suffices that user gathers required number od data on problem to be solved. The data may cover all information the user recognizes as being important. Neural net will behave like a black box with task to find relationships between inpu and output and to generalize its "private knowledge", so to be able to use it describe other similar process.
AitechSPHINX for Windows
PROJECT BY DR. KRZYSZTOF MICHALIK
|
Aitech Sphinx 
