Combining ELM with Random Projections for Low and High Dimensional Data Classification and Clustering

Extreme learning machine (ELM), as a new learning method for training feedforward neural networks, has shown its good generalization performance in regression and classification applications. Random projection (RP), as a simple and powerful technique for dimensionality reduction, is used for projecting high-dimensional data into low-dimensional subspaces while ensuring that the distances between data points are approximately preserved. This paper presents a systematic study on the application of RP in conjunction with ELM for both low-and high-dimensional data classification and clustering.