They propose a regularization method that changes the regularization factor for each data based on the Fisher information for data which has unbalanced and non-uniform variance of label noise . The accuracy is greatly improved when the degree of noise is different for each label.
TL;DR
They propose a regularization method that changes the regularization factor for each data based on the Fisher information for data which has unbalanced and non-uniform variance of label noise . The accuracy is greatly improved when the degree of noise is different for each label.
Why it matters:
Paper URL
https://arxiv.org/abs/2006.15766
Submission Dates(yyyy/mm/dd)
Authors and institutions
Methods
Results
Comments