Augmentation (Data Augmentation)

Data augmentation increases the variety of training examples without requiring entirely new data to be collected. For images, this may involve rotating, cropping, flipping, or adjusting brightness. For text, audio, and other data types, different transformations can be applied while maintaining the original meaning or classification. The augmented samples are then used alongside the original data during model training.

Machine learning models often perform better when trained on diverse datasets. Data augmentation helps increase this diversity, reduce overfitting, and improve a model's ability to generalize to unseen data. It is particularly valuable when collecting or labeling additional real-world data is expensive or difficult.