Comparison of the selection methods of neural network learning sample points

neural networks;  orthogonal design;  Uniform design

In order to compare the selection of sample points required for training artificial neural networks, the random traversal method, orthogonal design method and uniform design method are used to generate sample points for training neural networks. The analysis results show that under the same number of sample points, the uniform design method is the best representative, followed by the orthogonal design method, and the random traversal method is poor. The random traversal method can also improve its representativeness with the increase of the number of sample points. When the function varies less with the variable in the interval (less factor level can be taken), the orthogonal design method is also a good choice. The uniform design method can better reflect its superiority in the case of multiple variables and each variable needs to select a larger number of levels.