Generalized entropy based semi-supervised learning

Semi-supervised learning is a class of supervised learning techniques that also make use of unlabeled samples for training, the research aims to provide considerable improvement in learning accuracy with a small amount of labeled samples and affordable computational overhead. In this paper, we extend an probabilistic supervised learning model to semi-supervised multi-classification learning, both labeled and unlabeled samples are unified in our model levering the generalized entropy concept. For optimization, we adopt an efficient online learning algorithm which can achieve logarithmic regret with linear computational overhead in supervised learning situation. Empirical study shows our method obtain prediction accuracy closing to that of supervised learning while using extremely small labeled samples size.