Network Cooperation with Progressive Disambiguation for Partial Label Learning

Partial Label Learning (PLL) aims to train a classifier when each training instance is associated with set of candidate labels, among which only one correct but not accessible during the phase. The common strategy dealing such ambiguous labeling information disambiguate label sets. Nonetheless, existing methods ignore disambiguation difficulty instances and adopt single-trend mechanism. former would lead vulnerability models false positive labels latter may arouse error accumulation problem. To remedy these two drawbacks, this paper proposes novel approach termed "Network Cooperation Progressive Disambiguation" (NCPD) for PLL. Specifically, we devise progressive operations are performed on simple firstly then gradually more complicated ones. Therefore, negative impacts brought by can be effectively mitigated as ability model has been strengthened via learning from instances. Moreover, employing artificial neural networks backbone, utilize network cooperation mechanism trains collaboratively letting them interact other. As have different ability, interaction beneficial both reduce their respective errors, thus much better than algorithms process. Extensive experimental results various benchmark practical datasets demonstrate superiority our NCPD other state-of-the-art PLL methods.