Researchers at EPFL have developed a new, uniquely modular machine learning model for flexible decision-making. It is able to input any mode of text, video, image, sound, and time-series and then output any number, or combination, of predictions.