Researchers at the MIT-IBM Watson AI lab have developed a computationally efficient method that could be used to identify anomalies in the U.S. power grid in real time. The novel technique augments a special type of machine-learning model with a powerful graph structure, and does not require any labeled data to train.