Sequence Extreme Dynamic Online Learning

Extreme Learning Machine (ELM) is a novel learning algorithm with fast learning speed and good generalization performance through applying random mechanisms to assign and option the hidden layer parameter.Online sequen-tial ELM (OS-ELM) is an improved algorithm of ELM by using online learning scheme to overcome that ELM can not deal with large scale real-time data ef-fectively. To solve the difficult that Online Sequence Extreme Learning Machine could not change network structure once the learning process starts, the key point of this paper is how to chance network structure for OS-ELM algorithm The main work is presented as follows:Chapter1briefly introduces the basic concepts, network structure and de-velopment of artificial neural networks (ANNs) and the background and devel-opment of ELM algorithm.Chapter2systematically review the theory and procedure of ELM algo-rithm, two improved ELM algorithms (Dynamic ELM algorithm and Regularize ELM ELM algorithm)and OS-ELM algorithm.Chapter3develops two improved OS-ELM based on Dynamic ELM and Regularize ELM, where the Dynamic ELM can adjust network adaptively ac-cording to benchmark data sets what it learns and Regularize ELM can enhance the stable ability on adding regularize factor to compute output weight. Fur-thermore, the performance of the proposed algorithm Dynamic Online Sequence Extreme Learning Machine and Regularize Dynamic Online Sequence Extreme Learning Machine are tested by number value experiment.