QPSO-ELM: An evolutionary extreme learning machine based on quantum-behaved particle swarm optimization

Abstract

Extreme learning machine (ELM), as an emergent technology, has attracted tremendous attention from various fields for its fast learning speed. Different from traditional gradient-based learning algorithms for feed-forward neural networks, ELM need not be neuron alike and learns with good generalization performance. However, ELM may require more hidden neurons than traditional tuning-based learning algorithms in some applications due to the random assignment of the input weights and hidden biases. In this paper, a novel evolutionary ELM is proposed named QPSO-ELM which uses the quantum-behaved particle swarm optimization (QPSO) to select the input weights and hidden layer biases and reduces both the structural and empirical risks. The experimental results demonstrate the effectiveness of the proposed method with more compact networks.