or: Augment Your Batch: Improving Generalization Through Instance Repetition

On I/O-constrained training processes, it can be useful to re-use the same training example several times within a minibatch, with a different augmentation each time. Improves wall-clock time for reaching the same accuracy around convergence.