Thursday, April 14, 2022

On The Effects of Regularization and Data Augmentation are Class Dependent

Just read this paper from Facebook AI written by among others senior author Yann LeCun (Chief AI Scientist of Facebook).

This research paper describes a serious critique of two very common practices in machine/deep learning, i.e. data augmentation and weight decay.

However, I suspect a similar critique also applies to the imbalance of classes represented in the usual datasets.

[2204.03632] The Effects of Regularization and Data Augmentation are Class Dependent

No comments: