Determination Of Number Of ANNs’ Hidden Layer Neurons

Artificial neural network is an optimization model which can abstract and descriptthe human brain neural network from mathematics, physical and biological aspects.Because of ability of powerful mapping, it is widely used in various application fields.However, during the establishment of the artificial neural network model, one of thedifficulties is how to determine the number of hidden–layer-neurons.Firstly, the paper systematically introduces the artificial neural network model, andsummarizes the neural network model of hidden layer neuron number determinationmethod in domestic and foreign. Secondly, it is briefly discussed on some creativemethods, and analyzed the factors of the numbers of hidden layer neurons. Thirdly, byintroducing the polynomial and the thought of matrix pseudo inverse, the paper woulddiscuss the number of hidden layer neurons which are needed when the neural networkmodels are established in one dimension and multi-dimensional space respectively.In one dimension space, the paper would use the power function as hidden layerneuron activation function and combine with the thought of matrix inverse to firstlydetermine the neurons of the output weights and establish the polynomial neuralnetwork models. Finally, Through the introduction of interval binary search method, thebest number of the hidden layer neurons would be determined.In the multi-dimensional space, the paper uses the specific way to determine theinput weights so that the purpose of dimension conversion would be realized, and usesthe S function as the neuron activation function, combines with the thought of matrixinverse, and firstly determines the neurons of the output weights, the polynomial neuralnetwork model would be established at last. And according to the dimension of thespace of polynomials, the number of the hidden layer neurons of the neural networkmodel would be worked out.