Label Smoothing as Another Regularization Trick

What is label smoothing and how to implement it in PyTorch

Photo by Dave on Unsplash

Overfitting and probability calibration are two issues that arise when training deep learning models. There are a lot of regularization techniques in deep learning to address overfitting; weight decay, early stopping, and dropout are some of the most popular ones. On the other hand…