Semi-supervised learning method of constructive neural networks

Constructive Machine Learning algorithm is widely used in supervised learning for its advantages of low computing complexity and high training speed. However, when the labeled training samples are not enough, the supervised learning quality is not ideal enough, hence its application domain is limited. A semi-supervised learning method is proposed to solve this problem. The subordinate degree of unlabeled sample and the classifying believe degreed are set up. On this basis, the self-training strategy is utilized to conduct semi-supervised learning. The semi-supervised learning method improves the performance of classifier and broadens the application domain. The experimental results demonstrate the validity of the method.