Multi-parallel Extreme Learning Machine with Excitatory and Inhibitory Neurons for Regression

摘要

Compared with traditional neural networks, extreme learning machine (ELM) shows outstanding performances on speed and computation. Aiming at the problems that ELM needs more hidden layer neurons and meaningful features of data sometimes are sacrificed in order to improve the training speed, a novelty network multi-parallel extreme learning machine with excitatory and inhibitory neurons (MEI-ELM) is proposed based on the idea of biological neurons. In MEI-ELM, (1) A parallel system is introduced to make it more compact and reduce the number of hidden layer neurons. (2) The property of excitatory and inhibitory of biological neuronal for data processing is introduced to improve its performance. Through applying MEI-ELM, ELM, Fast Learning Network (FLN) and Fast Learning Network with Parallel Layer Perceptrons (PLP-FLN) to 11 classical regression problems, it can be obtained that MEI-ELM performs much better than the other methods in generalization and stability.