| Abstrakt: |
At present, in typical architectures of feedforward neural networks, the values of the weights of the connections and thresholds of neurons are determined by adjusting the values of the weights, performed by means of typical learning algorithms. The architectures of feedforward neural networks implemented on the basis of metric recognition methods are also known, the values of the weights of neurons for which are precalculated analytically. The analytical calculation of the weight values is carried out on the basis of metric expressions and allows a workable neural network to be immediately obtained without training. In this case, the effectiveness of the obtained neural network depends on the selected set and the number of samples, as well as on the selected dimension of the table of weights. Such neural networks can also be trained with typical learning algorithms, which makes it possible to increase the efficiency of the neural network with the calculated weights through additional training of the neural network. Here, the process of calculating the weight values and the further training of the neural network is also faster than training the neural network in the traditional way; on the basis of these networks, the possibility of determining the weight values and thresholds of a neural network using the strength and potential of the electrostatic field is considered. That is, it is proposed to use the parameter values of the electrostatic field as weight values of a neural network. In other words, the possibility of creating a workable neural network without analytical calculations and without the use of learning algorithms is considered. This approach allows the process of determining the values of the neural network weights almost instantaneous. The technically possible implementations of this approach and the problematic aspects of using the parameters of the electrostatic field as weights of a neural network, as well as possible approaches to resolving these difficulties are considered. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink