A fast RBM-hidden-nodes based extreme learning machine
Li Chen, Ling Yang, Chao Sun, Haipeng XI · 2017
In this paper, we propose an Extreme Learning Machine (ELM) approach for solving large and complex data problems. In contrast to existing approaches, we embed hidden nodes that are designed using Restricted Boltzmann machine (RBM) into the classical ELM, exhibiting excellent generalization performances. To overcome the high computational complexity involved especially on large datasets, hidden nodes are derived from RBM trained in turn by multiple random subsets of data sampled from the original datasets instead of the entire dataset in one time. The resultant algorithm proposed is labeled here as FRBM-H-ELM in short. Comprehensive experiments and comparisons are conducted to assess the FRBM-H-ELM against the traditional Extreme Learning Machine. The results obtained demonstrated the superior generalization performance and efficiency of FRBM-H-ELM.