Our lab for Efficient Learning and Probabilistic Inference for Science (ELPIS) at Helmholtz AI and TU Munich studies the interface between Bayesian inference and deep learning with the goals of improving robustness, data-efficiency, and uncertainty estimation in modern machine learning approaches.

While deep learning often leads to impressive performance in many applications, it can be over-confident in its predictions and require large datasets to train. Especially in scientific applications, where training data is scarce and detailed prior knowledge is available, insights from Bayesian statistics can be used to drastically improve these models.

Important research questions include how to effectively specify priors in deep Bayesian models, how to harness unlabeled data to learn re-usable representations, how to transfer knowledge between tasks using meta-learning, and how to guarantee generalization performance using PAC-Bayesian bounds.

The mission statement of our group is to make Bayesian deep learning a viable default solution for all scientific learning tasks. This will require foundational research on Bayesian machine learning, such as regarding priors and inference, applied research to make the existing methods more usable in practice, and collaborative research with domain scientists to keep our solutions grounded in their actual problems.

Our three core values are Bayesianism, both to inspire better machine learning methods and as an approach to update our beliefs based on new evidence during our research process, curiosity, to drive our research forward and motivate us not to stop our investigations before getting to the core of the questions, and hope, both regarding the success of each new research project and the overall impact that our research will have on the world. Note that the latter is also reflected in our group’s name, ἐλπίς (elpis) being the Ancient Greek spirit of hope.