Interpolation Consistency Training for Semi-Supervised Learning
Abstract
We introduce Interpolation Consistency Training (ICT), a simple and computation efficient algorithm for training Deep Neural Networks in the semi-supervised learning paradigm. ICT encourages the prediction at an interpolation of unlabeled points to be consistent with the interpolation of the predictions at those points. In classification problems, ICT moves the decision boundary to low-density regions of the data distribution. Our experiments show that ICT achieves state-of-the-art performance when applied to standard neural network architectures on the CIFAR-10 and SVHN benchmark dataset.
Cite
Text
Verma et al. "Interpolation Consistency Training for Semi-Supervised Learning." International Joint Conference on Artificial Intelligence, 2019. doi:10.24963/IJCAI.2019/504Markdown
[Verma et al. "Interpolation Consistency Training for Semi-Supervised Learning." International Joint Conference on Artificial Intelligence, 2019.](https://mlanthology.org/ijcai/2019/verma2019ijcai-interpolation/) doi:10.24963/IJCAI.2019/504BibTeX
@inproceedings{verma2019ijcai-interpolation,
title = {{Interpolation Consistency Training for Semi-Supervised Learning}},
author = {Verma, Vikas and Lamb, Alex and Kannala, Juho and Bengio, Yoshua and Lopez-Paz, David},
booktitle = {International Joint Conference on Artificial Intelligence},
year = {2019},
pages = {3635-3641},
doi = {10.24963/IJCAI.2019/504},
url = {https://mlanthology.org/ijcai/2019/verma2019ijcai-interpolation/}
}