Abstract:

A model of semi-supervised classification was proposed to overcome the problem induced by mislabeled samples. A decision rule was learned from labeled and unlabeled data,and a new energy function based on robust error function was used in the Markov random field. Also two algorithms based on the iterative condition mode and the Markov chain Monte Carlo were designed to infer the label of both labeled and unlabeled samples. Experimental results demonstrated that the proposed methods were efficient for a real-world dataset.

Key words:  semi-supervised; classification; Markov random field; robust error function