13.07.2015 Views

Neural network software tool development in C

Neural network software tool development in C

Neural network software tool development in C

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Figure 15: New neural <strong>network</strong> controlThe features on this neural <strong>network</strong> library that are more important to our <strong>software</strong> <strong>tool</strong> are describedbelow:1. Ability to solve problems of function approximation and classification;2. The classification problem is a problem of separat<strong>in</strong>g two classes of po<strong>in</strong>ts;3. The data is <strong>in</strong>putted to the neural <strong>network</strong> by click<strong>in</strong>g <strong>in</strong> a white panel ( <strong>in</strong> classification problemsthis po<strong>in</strong>t has different colors that are displayed if the user clicks on the right or left mousebuttons);4. Four activation functions:• Sigmoid;• L<strong>in</strong>ear;• Heaviside;• Gaussiana.5. All weights are <strong>in</strong>itialized, by default, to zero;6. The <strong>in</strong>itial weights can be altered by generat<strong>in</strong>g random values <strong>in</strong> the <strong>in</strong>terval [−1, 1]. This<strong>in</strong>terval is set by default but can be altered by the user;7. The <strong>network</strong> is fully connected;8. By default, the activation function is Sigmoid and this activation function can be different foreach neuron or each layer;9. The first layer is a process<strong>in</strong>g units;13

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!