1 paper
Sachin Chhabra, Hemanth Venkateswara, Baoxin Li
Training neural networks with one-hot target labels often results in overconfidence and overfitting. Label smoothing addresses this issue by perturbing the one-hot target labels by…