科研成果详情

We introduce a statistical procedure for the simultaneous clustering and feature selection of positive vectors. The proposed method is based on well-principled infinite generalized inverted Dirichlet (GID) mixture models. Unlike most currently existing learning algorithms, based on frequentist or Bayesian approaches, that have been proposed for GID mixtures, our framework is based on stochastic expectation propagation that offers a good compromise. The performance of the proposed algorithm is examined using both simulated and real world data sets.