Researchers at MIT and the University of Basel have developed a physics-informed machine-learning approach to automatically map out phase diagrams for novel physical systems. This approach is more efficient and does not require large labeled training datasets.