To provide fast and reliable scientific prediction, an ideal ML workflow is expected to achieve high prediction accuracy while requiring minimal experimental costs for training and inference. We ...