|
|
|
     |
|
|
|
|
|
|
|
| |||||||||
|
|
Intelligent Input Neurons Input neurons are on the whole simply data dispatchers, they capture the data
receiving at the network and pass them over to the succeeding neurons,
regardless of their magnitude and bias. Since the succeeding
neurons require the data in a particular shape - e.g. to prevent from operating
in the saturated zone - the data need to be normalized prior to
learning.
To prevent from normalisation prior to training especially with respect to the generally unknown optimal method, that part is integrated into the learning process of the neural network and yields in combination with the heuristic network initialisation an optimal shape for the network weights. Intelligent input neurons can be considered as neurons that are promoted to participate at the learning process instead of being only simple data dispatcher. The normalization of the input data is more sophisticated since it is dynamic, the normalization parameters are optimised throughout the whole learning process, yielding faster and more accurate results, no more pre-processing with respect to normalisation of the data is necessary. Intelligent input neurons are the default in cVision. |
|