Extreme Learning Machine and Its Applications in Big Data Processing

Abstract The extreme learning machine (ELM) is widely used in batch learning, sequential learning, and incremental learning because of its fast and efficient learning speed, fast convergence, good generalization ability, and ease of implementation. With the development of the traditional ELM, lots of improved ELM algorithms have been proposed; meanwhile the scope of implementing the ELM has been further expanded from supervised learning, to semisupervised learning and unsupervised learning. However, due to its memory-residency, and high space and time complexity, the traditional ELM is not able to train big data fast and efficiently. Optimization strategies have been employed for the traditional ELM to solve this problem. In this chapter, we will first review ELM theories and some important variants, and then describe parallel ELM algorithms based on MapReduce and Spark in detail. Lastly, we show some practical applications of the ELM for big data.