Multi-layer Online Sequential Extreme Learning Machine for Image Classification

Abstract

In this paper, a multi-layer online sequential extreme learning machine (ML-OSELM) is proposed for image classification. ML-OSELM is an online sequential version of a recently proposed multi-layer extreme learning machine (ML-ELM) method for batch learning. Existing ELM-based sequential learning methods, such as state-of-the-art online sequential extreme learning machine (OS-ELM), were proposed only for single-hidden-layer networks. A distinctive feature of the new method is that it can sequentially train a multi-hidden-layer ELM network. Auto-encoders are used to perform layer-by-layer unsupervised sequential learning in ML-OSELM. We used four image classification datasets in our experiments and ML-OSELM performs better than the OS-ELM method on all of them.