What is label smoothing and why is it used in classification models?

Label smoothing softens hard labels to prevent overconfident predictions.

Label smoothing replaces hard one-hot labels (0 and 1) with softened probabilities (e.g., 0.9 and 0.1). This prevents the model from becoming overly confident and improves generalization. It also reduces overfitting and helps neural networks produce better-calibrated probabilities. It is widely used in modern deep learning models like transformers.