Estimation of the optimal number of neurons in extreme learning machine using simulated annealing and the golden section

Abstract Extreme learning machine is a neural network algorithm widely accepted in the scientific community due to simplicity of model and its good results classification regression problems; digital image processing, medical diagnosis, signal recognition are some applications field physics addressed with these networks. The must be executed an adequate number neurons hidden layer obtain results. Identifying appropriate open problem extreme field. search process has high computational cost if carried out sequentially, given complexity calculations as increases. In this work, we use golden section simulated annealing heuristic methods calculate Learning Machine; for experiments, three real databases were used synthetic database problem. show that accelerated up 4.5× times 95.7× compared sequential method highest-dimensional database.