GM-PLL: Graph Matching Based Partial Label Learning

Partial Label Learning (PLL) aims to learn from the data where each training example is associated with a set of candidate labels, among which only one correct. The key deal such problem disambiguate label sets and obtain correct assignments between instances their labels. In this paper, we interpret as instance-to-label matchings, reformulate task PLL matching selection problem. To model problem, propose novel Graph Matching based (GM-PLL) framework, (GM) scheme incorporated owing its excellent capability exploiting instance relationship. Meanwhile, since conventional one-to-one GM algorithm does not satisfy constraint that multiple may correspond same label, extend traditional probabilistic many-to-one constraint, make proposed framework accommodate Moreover, also relaxed prediction model, can improve accuracy via strategy. Extensive experiments on both artificial real-world demonstrate method achieve superior or comparable performance against state-of-the-art methods.