Kernel Extreme Learning Machine Based On Conjugate Gradient And Manifold Regularization And Its Application

Extreme learning machine(ELM)is an improved model of feedforward neural network.Because of its simple parameter setting and fast learning speed,it has a wide range of applications and related algorithms.The extreme learning machine obtains the hidden layer random mapping by systematically giving the input layer weights.The hidden layer weights are obtained by one calculation using the least squares method,which makes the learning speed very fast.In this paper,the extreme learning machine algorithm is researched,and some defects of the extreme learning machine are improved.Two new extreme learning machine models are proposed and applied to image restoration and diabetes detection,which improves the performance of the extreme learning machine and broadens the application of extreme learning machines.The main works of this paper are as follows:(1)In order to improve the stability of the extreme learning machine network and improve its generalization performance,the regular extreme learning machine finally reconciles the ratio of empirical risk to structural risk by introducing regularization theory and combining statistical risk knowledge.In this paper,the regular extreme learning machine is used to recover the image affected by Gaussian blur.Compared with the BP network and the mean filtering restoration algorithm,it has a good performance in training speed and recovery effect.(2)Based on the introduction of the kernel function theory,the kernel extreme learning machine uses the kernel map as the hidden layer output,which has better stability and shows superiority over random mapping.However,when the sample is too large,the method of obtaining the output weight by solving the inverse matrix by the kernel extreme learning machine does not perform well in time and space complexity.Therefore,this paper proposes a kernel extreme learning machine based on conjugate gradient.By introducing a conjugate gradient algorithm to solve the inverse matrix in the kernel extreme learning machine,it can have an advantage in learning speed,and the experiment shows that the occupied memory is reduced.Better viability on the computer.The improved kernel extreme learning machine is used to recover Gaussian blurred images,and compared with BP network and mean filtering.The results show that the proposed improved nuclear limit learning machine has advantages in terms of time and effect.(3)In this paper,a popular regularized kernel extreme learning machine model is proposed,and the manifold regularization theory is introduced.The Laplace matrix method is used to improve the learning algorithm of the nuclear extreme learning machine.The simulation experiment verifies the good generalization performance of the manifold regular kernel extreme learning machine.Although it increases the time consumption,it still performs well in effect,which is a new method.