If you take the following sentence from an article on deep neural networks
to regularize the classifier layer by estimating the marginalized effect of label-dropout during training.
What does label-dropout mean
Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
Sign up to join this communityIf you take the following sentence from an article on deep neural networks
to regularize the classifier layer by estimating the marginalized effect of label-dropout during training.
What does label-dropout mean
I ended up finding the paper you were referencing, maybe next time, add it to your post.
From what I can understand in this paper, label-dropout means that you are dropping the real labels and replace them with others.
The section of the paper you're referring to is explaining everything in great detail, so I'd try to read it carefully. In short, instead of using the ground-truth distribution (i.e. one 1 and a bunch of 0s) in the target layer, they are adding some uncertainty. That way, the model becomes less confident, and therefore is less prone to overfitting and becomes more adaptable.
Perhaps it means groundtruth just keep one correct label while dropping out others. And this can cause some side-effects under extreme circumstances.