A Convex Formulation for Learning from Crowds

Recently crowdsourcing services are often used to collect a large amount of labeled data for machine learning, since they provide us an easy way get labels at very low cost and in short period. The use has introduced new challenge that is, coping with the variable quality crowd-generated data. Although there have been many recent attempts address problem multiple workers, only few existing methods consider learning classifiers directly from such noisy All these modeled true as latent variables, which resulted non-convex optimization problems. In this paper, we propose convex formulation crowds without estimating by introducing personal models individual crowd workers. We also devise efficient iterative method solving problems exploiting conditional independence structures classifiers. evaluate proposed against three competing on synthetic sets real crowdsourced set demonstrate outperforms other methods.