Behavior-learning based semi-supervised kernel extreme learning machine for classification

Semi-supervised Extreme Learning Machine is proved to be an efficient algorithm for multiple semi-supervised classification tasks. However, the weights connecting the input layer and the hidden layer, and the bias vector of the hidden layer are generated randomly, making the classifier unstable. Besides, an improper choice of heat kernel parameter will badly influence the performance of the classifier when constructing the Laplacian matrix. In this paper, we introduced kernel theory to eliminate the uncertainty of input weights and adopted human behavior learning strategy to make it easier to find proper heat-kernel parameters. The simulation experiments show that the behavior-learning based semisupervised extreme learning machine achieves a better classification performance compared with Semisupervised Extreme Learning Machine.