ELM*: distributed extreme learning machine with MapReduce

Extreme Learning Machine (ELM) has been widely used in many fields such as text classification, image recognition and bioinformatics, as it provides good generalization performance at a extremely fast learning speed. However, as the data volume in real-world applications becomes larger and larger, the traditional centralized ELM cannot learn such massive data efficiently. Therefore, in this paper, we propose a novel Distributed Extreme Learning Machine based on MapReduce framework, named ELM (auaEuro parts per thousand), which can cover the shortage of traditional ELM whose learning ability is weak to huge dataset. Firstly, after adequately analyzing the property of traditional ELM, it can be found out that the most expensive computation part of the matrix Moore-Penrose generalized inverse operator in the output weight vector calculation is the matrix multiplication operator. Then, as the matrix multiplication operator is decomposable, a Distributed Extreme Learning Machine (ELM (auaEuro parts per thousand)) based on MapReduce framework can be developed, which can first calculate the matrix multiplication effectively with MapReduce in parallel, and then calculate the corresponding output weight vector with centralized computing. Therefore, the learning of massive data can be made effectively. Finally, we conduct extensive experiments on synthetic data to verify the effectiveness and efficiency of our proposed ELM (auaEuro parts per thousand) in learning massive data with various experimental settings.