In Semi-Supervised Semi-Private (SP) learning, the learner has access to both public unlabelled and private labelled data. We propose PILLAR, an easy-to-implement and computationally efficient ...