Representational Learning with Extreme Learning Machine for Big Data Liyanaarachchi

Restricted Boltzmann Machines (RBM) and auto encoders, learns to represent features in a dataset meaningfully and used as the basic building blocks to create deep networks. This paper introduces Extreme Learning Machine based Auto Encoder (ELM-AE), which learns feature representations using singular values and is used as the basic building block for Multi Layer Extreme Learning Machine (ML-ELM). ML-ELM performance is better than auto encoders based deep networks and Deep Belief Networks (DBN), while in par with Deep Boltzmann Machines (DBM) for MNIST dataset. However MLELM is significantly faster than any state−of−the−art deep networks.