Universum Prescription: Regularization Using Unlabeled Data

By Yann LeCun. 2015

33 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xin Li
2. Fuxin Li
3. Nicholas Carlini
4. Cristian Sminchisescu
5. David A. Wagner
6. João Carreira

In Computer Vision and Pattern Recognition

1. Xin Li
2. Fuxin Li
3. Nicholas Carlini
4. Jitendra Malik
5. Cristian Sminchisescu
6. João Carreira
7. Luc Van Gool
8. Ming-Hsuan Yang

Citations in the corpus

• Adversarial Examples Detection in Deep Networks with Convolutional Filter Statistics (2016, background): “The universum classifier [31] is similar but with more theoretical arguments.”