Universum Prescription: Regularization using Unlabeled Data
This provides a simple method for enhancing deep learning performance with unlabeled data, but it is incremental as it builds on existing regularization approaches.
The paper tackles the problem of improving supervised learning by using unlabeled data as a regularization technique, achieving competitive results on CIFAR-10, CIFAR-100, STL-10, and ImageNet datasets.
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.