Abstract: The present manuscript addresses the question of how training data should be sampled for deep learning models by analyzing and evaluating the impact of training data representation and ...