EPFL researchers have developed a new modular machine learning model, MultiModN, that can learn from various inputs and output multiple predictions. This model aims to address the issue of biased missing information and is presented at the NeurIPS2023 conference.