YIP+ Learning from learned representations

Learning from learned representations 

Machine learning has the potential to guide experimental sciences through the jungle of unperformed experiments to focus on the most informative ones. Current models can often capture the inner workings of complex phenomena, but they are hard for humans to learn from. This limits their use when forming new hypotheses about the observed phenomenon. This project builds geometric tools for interpreting and learning from density estimating models. The project hires one PhD student and one postdoc.