Here comes the Part 3 on learning with not enough data (Previous: Part 1 and Part 2). Let’s consider two approaches for generating synthetic data for training.
Augmented data. Given a set of existing training samples, we can apply a variety of augmentation, distortion and transfor...