The Neuro Control Manager:
Always One Step Ahead!

The Neuro Control Manager (NeuroCoM) is a high performance instrument for delveloping and testing of neural networks. Running under MS-Windows, its window-oriented GUI facilitates the input of any parameter for both, the analysis and the training of neural networks. Furthermore, the GUI visualizes the structure, the course of the learning process and the transfer function by vivid graphics. For the implementation of any neural network to any type of target hardware the NeuroCoM provides the output of C source code as well as the export of the neurons' weighting/offset data.

Main Features of the NeuroCoM:

Configuring neural networks:

• The number of inputs and outputs may be specified according to the respective goal.

• A definite number of hidden layers and the number of neurons per layer may be selected dependent on the complexity of the transformation to be realized.

• The generation of start values may be influenced by specifying the evaluation range and by initializing the random number generator.

Control of the Learning Process:

• Three different learning methods may be selected (standard error backpropagation, adaptive backpropagation and quickprop).

• By the introduction of a learning factor 'Eta' the extent of the error backpropagation and, thus, the relation of learning velocity and accuracy of the learning process may be influenced.

• A modificable integration factor (momentum) determines the consideration of the results of preceding learning steps.

• Further optimizing options are available by selecting "flat spot elimination" or "output error distortion".

• The NeuroCoM just reads the real learning data (input data and demanded output values) from a simple structurized ASCII file.

• The training process is automatically aborted when the total errors falls below the specified threshold or when the defined maximum of learning steps is reached. In addition, it may, of course, be aborted directly by the user.

Feasiblities of Visualization and Testing:

• During the training modifications of the structure of the neural network may be monitored by means of the graphic.

• The learning behaviour of the network is indicated by a graphic curve of error which, in addition, gives information on the suitability of the learning parameters selected.

• The transfer function of the neural network is visualized by 3D graphics (also available during training).

• For testing the trained network exemplary sets of data may be input from a file and the results may be compared to the given output values. The analysis may be performed individually for each test data set or statistically for the entire test data file.

Screen shot · Softcomputing · Contents