Research

I am interested in methods that let us infer scientific knowledge from data. My main focus is simulation-based (or likelihood-free) inference: statistical methods for the common case when we can model a phenomenon with a computer simulation, but we cannot calculate its likelihood function. We have pioneered new techniques that combine some understanding of the latent processes in the simulator, a dash of classical statistics, and a heavy dose of machine learning.

In particle physics, these methods allow us to measure fundamental physics properties more precisely than before. I am the lead developer of the MadMiner library, which makes it straightforward to apply these algorithms to almost any LHC problem. I also worked on forecasting methods based on information geometry for particle physics and studied the phenomenology of effective field theories, a language that lets us characterize for instance the properties of the Higgs boson.

We generalized some of the methods that we originally developed for particle physics and applied them to a very different problem: to learn about dark matter properties from observations of strong gravitational lensing. In fact, these new ideas are not at all limited to physics, and I am excited to find out how they can help us understand what is happening in scientific fields ranging from economics to epidemiology.

Beyond scientific applications, I am broadly interested in (approximate) inference, normalizing flows, and generative models.

Publications

For my papers, see my INSPIRE profile or Google Scholar.

Most of my scientific code is on my GitHub repo. The repository also contains slides from most of my talks.

Contact

Find me, most of the time, in room 840 in the Department of Physics, or in room 614 at the Center for Data Science. Or write me at johann.brehmer@nyu.edu.

What else?

I sometimes travel and take pictures: