The Research And Application Based On Extreme Learning Machine

Neural network make up a large proportion in the field of Machine learning and pattern recognition. The traditional Error Back Propagation neural network and Support Vector Machine exist some problem such as easy to fall into local minima, slow convergence speed and over fitting. Because of the efficient fast and good generalization performance characteristics, the Extreme Learning Machine has got the attention, the research and application by many scholars at home and abroad.With the development of information technology, big data has become one of the hot spot of social research. The progress of science and technology make people obtained these data which present the complex structure, huge number, more dimension features. These data has timeliness so we need process and analysis the data timely to get the information and value behind it. Therefore it’s full of the challenges to process and analysis the data. And it’s especially important to visual the high-dimensional data.Based on the recent popular Extreme Learning Machine neural network, use MDS, Pearson correlation and Spearman correlation respectively instead of the common mean square error to change the updating rule of the network from different angles to project data visualization about the high-dimensional data onto a two-dimensional plane. Compared the proposed method with SNE and t-SNE which have been popular recently, experimental results show that visual effect and computational performance for the proposed methods significantly better than the SNE and t-SNE. Furthermore, these experimental results also show that the MDS-based ELM exhibits the best performance.On image recognition, the current popular algorithm mainly adopts convolutional neural network for feature extraction and classification. This method achieved good effect usually on huge and complex neural network. The fully connected layers of convolutional neural network formed the classifier which uses the gradient descent method to train, therefore the generalization ability has been limited. Convolution neural network has enough discrimination, and extreme learning machine has good generalization performance. So use the convolutional neural network for image feature extraction, and use extreme learning machine as a classifier, to establish a model of image recognition based on CNN-ELM. The experimental results show that compared with the classical model of CNN, the proposed model’s recognition rate improved obviously.