In semi-supervised learning, the weights of unlabeled data are usually treated uniformly, but they propose a method for automatically determining the weights of individual data, which can be incorporated into existing loss systems such as FixMatch and significantly improve the accuracy.
TL;DR
In semi-supervised learning, the weights of unlabeled data are usually treated uniformly, but they propose a method for automatically determining the weights of individual data, which can be incorporated into existing loss systems such as FixMatch and significantly improve the accuracy.
Why it matters:
Paper URL
https://arxiv.org/abs/2007.01293
Submission Dates(yyyy/mm/dd)
2020/07/02
Authors and institutions
Zhongzheng Ren, Raymond A. Yeh, Alexander G. Schwing
Methods
After several updates using semi-supervised losses, use Hessian to update the individual weights of the data.
Results
Improved on CIFAR10, SVHN
Comments