Variational Gaussian process for missing label crowdsourcing classification problems

In this paper we address the crowdsourcing problem, where a classifier must be trained without knowing the real labels. For each sample, labels (which may not be the same) are provided by different annotators (usually with different degrees of expertise). The problem is formulated using Bayesian modeling, and considers scenarios where each annotator may label a subset of the training set samples only. Although Bayesian approaches have been previously proposed in the literature, we introduce Variational Bayes inference to develop an iterative algorithm where all latent variables are automatically estimated. In the experimental section the proposed model is evaluated and compared with other state-of-the-art methods on two real datasets.