Multitask Learning Using Nonparametrically Learned Predictor Subspaces

Given several related learning tasks, we propose a nonparametric Bayesian learning model that captures task relatedness by assuming that the task parameters (i.e., weight vectors) share a latent sub