Improved Mixed-Example Data Augmentation

Cecilia Summers, Michael J. Dinneen

In order to reduce overfitting, neural networks are typically trained with data augmentation, the practice of artificially generating additional training data via label-preserving…