Learning internal representations from data using no or few labels is useful\nfor machine learning research, as it allows using massive amounts of unlabeled\ndata. In this work, we use the Bayesian Confidence Propagation Neural Network\n(BCPNN) model developed as a biologically plausible model of the cortex. Recent\nwork has demonstrated that these networks can learn useful internal\nrepresentations from data using local Bayesian-Hebbian learning rules. In this\nwork, we show how such representations can be leveraged in a semi-supervised\nsetting by introducing and comparing different classifiers. We also evaluate\nand compare such networks with other popular semi-supervised classifiers.\n