Proposed a new kind of improvement BP algorithm based on sample expected training number.

With the Kohonen network clustering in neural network employed, the degree of relationship of the universal joint axle of the rolling mill was input to Kohonen network as the training sample, studied and clustered by the network to generate different clustering centers according to the different depth and different degree of relationship among the cracks.

Exact agricultural crops identification and planting area measure depend on not only classifiers but also training samples imported into classifiers.

Typical fault characteristics are selected as training samples and GA is used to optimize the structure and original weight distribution of BP networks.

Studies on the purification of training samples in supervised classification by mode filtering;

A purified algorithm for training samples based on local automatic searching and spectral matching technique is presented.

In this paper, using orthogonal experiment method, taking three layers feed forward neural network as a example, problem of choosing training samples, weights and bias, training parameter of neural network is analyzed and studied.

In this paper, relevant parameters influencing sample training precision are explored when the mixtures of experts networks is applied to boiler fault diagnosis,and the relationahip between the square sum of studying error and some parameters,such as rule numbers、studying rate、cycle numbers、weighted exponent,are presented.

The statistical learning theory based on sample training was applied to analyze the difference and relativity of images figures,and then the common figures and distinct figures of each sort image were catched to form the identification and classification model.

A new method based on Discriminative Common Vector in case of one training sample scenario was presented.