Extreme learning machine with sparse connections

The thesis is in the field of machine learning, and specifically studies the recent emerging algorithm, Extreme Learning Machine (ELM). Unlike previous ELM implementations, in which hidden nodes are in full connection with the input ones, we present the ELM with sparse connections. In one way, it reduces the storage space and testing time, while providing better scalability for large-scale applications. In the other way, the sparse connections make it especially suitable and efficient for locally correlated applications, such as image processing, speech recognition, etc.DOCTOR OF PHILOSOPHY (EEE