A Semi-supervised Incremental Learning Algorithm Based on Auto-adaptive Probabilistic Hypergraph and Its Application for Video Semantic Detection

Abstract

Semantic categorization of complex videos is an ambiguous task. The semi-supervised learning method, which is based on hyper graph model, can achieve multi-semantics labels, but it is sensitive to the radius parameter when a hyper graph model is constructed and the number of vertices belonging to a hyper edge is fixed. A new method is proposed in this paper to construct an auto-adaptive probabilistic hyper graph (ada-PHGraph) model, where a formula is presented as a measurement to auto-adaptively decide whether a vertex is belonged to a hyper edge or not. Our proposed algorithm has high robustness and can overcome the defect of fixed number of vertices belonging to the same hyper edge in the traditional probabilistic hyper graph model. In addition, a pre-defined threshold is used to judge whether the model learning result for unlabeled samples has high certainty and can been included in the model. The auto-adaptive probabilistic hyper graph model can achieve the dynamic updates effectively when the number of samples increases by applying the incremental learning mechanism. Our experimental results have shown that the auto-adaptive probabilistic hyper graph model can improve the model generalization ability and utilize the unlabeled samples effectively.