Universum Prescription: Regularization using Unlabeled Data

Abstract

This paper shows that simply prescribing "none of the above" labels to unlabeled data has a beneficial regularization effect to supervised learning. We call it universum prescription by the fact that the prescribed labels cannot be one of the supervised labels. In spite of its simplicity, universum prescription obtained competitive results in training deep convolutional networks for CIFAR-10, CIFAR-100, STL-10 and ImageNet datasets. A qualitative justification of these approaches using Rademacher complexity is presented. The effect of a regularization parameter -- probability of sampling from unlabeled data -- is also studied empirically.

7 pages for article, 3 pages for supplemental material. To appear in AAAI-17

References in corpus (6)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Distilling the Knowledge in a Neural Network
• Theoretical Models of Learning to Learn
• Stochastic Pooling for Regularization of Deep Convolutional Neural Networks
• Stacked What-Where Auto-encoders
• Spatially-sparse convolutional neural networks