Research

I am interested in methods that let us infer knowledge from data. The main focus of my research is simulation-based (or likelihood-free) inference: statistical methods for the case when we can model a phenomenon with a computer simulation, but we cannot calculate its likelihood function. This scenario is very common in scienctific fields from neuroscience to epidemiology and from elementary particle physics to cosmology. We have pioneered new inference techniques that combine some understanding of the latent processes in the simulator, a dash of classical statistics, and a heavy dose of machine learning.

Applied to problems 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 measurement problem at the LHC experiments. I also worked on forecasting methods based on information geometry for particle physics.

But these methods are not limited to particle physics. We used the same techniques to learn about dark matter through gravitational lensing. I am excited to find out how they can help us understand what is happening in problems in many other scientific domains.

Beyond scientific use cases, I am generally interested in (approximate) inference, normalizing flows, and generative models. I have led the devlopment of manifold-learning flows, a new class of generative models that simultaneously learn the data manifold as well as a tractable probability density on that manifold.

Publications

For a full list of my publication, see my INSPIRE profile or Google Scholar.

I would like to highlight the following papers:

  • JB and Kyle Cranmer:
    Flows for simultaneous manifold learning and density estimation.
    [arXiv] [bibtex] [code]
  • JB, Kyle Cranmer, Siddharth Mishra-Sharma, Felix Kling, and Gilles Louppe:
    Mining gold: Improving simulation-based inference with latent information.
    NeurIPS workshop on Machine Learning and the Physical Sciences (2019). [workshop]
  • Kyle Cranmer, JB, and Gilles Louppe:
    The frontier of simulation-based inference.
    Accepted by the Proceedings of the National Academy of Science. [arXiv] [bibtex]
  • JB, Siddharth Mishra-Sharma, Joeri Hermans, Gilles Louppe, and Kyle Cranmer:
    Mining for Dark Matter Substructure: Inferring subhalo population properties from strong lenses with machine learning.
    The Astrophysical Journal 886 (2019). [arXiv] [journal] [bibtex] [code]
  • JB, Felix Kling, Irina Espejo, and Kyle Cranmer:
    MadMiner: Machine learning-based inference for particle physics.
    Computing and Software for Big Science 4 (2020). [arXiv] [journal] [bibtex] [code]
  • Markus Stoye, JB, Gilles Louppe, Juan Pavez, and Kyle Cranmer:
    Likelihood-free inference with an improved cross-entropy estimator.
    NeurIPS workshop on Machine Learning and the Physical Sciences (2019). [arXiv] [workshop]
  • JB, Gilles Louppe, Juan Pavez, and Kyle Cranmer:
    Mining gold from implicit models to improve likelihood-free inference.
    Proceedings of the National Academy of Science 117 (2020). [arXiv] [journal] [bibtex]
  • JB, Kyle Cranmer, Gilles Louppe, and Juan Pavez:
    Constraining Effective Field Theories with Machine Learning.
    Physical Review Letters 121 (2018). [arXiv] [journal] [bibtex] [code]
  • JB, Kyle Cranmer, Felix Kling, Tilman Plehn:
    Better Higgs Measurements Through Information Geometry.
    Physical Review D 95 (2017). [arXiv] [journal] [bibtex]
  • JB, Ayres Freitas, David Lopez-Val, Tilman Plehn:
    Pushing Higgs Effective Theory to its Limits.
    Physical Review D 93 (2016). [arXiv] [journal] [bibtex]

Code

Together with Felix Kling, Irina Espejo, and Kyle Cranmer, I have developed the MadMiner library, which automates machine learning–powered techniques for simulation-based inference for particle physics experiments.

You may also be interested in my GitHub repository, which contains the code used for most of my papers, for instance the code for manifold-learning flows.

Talks

My Github repository also contains PDF versions of slides from most of my talks.

Contact

Write me at johann.brehmer@nyu.edu. In non-pandemic times, you can usually find me either in room 614 at the Center for Data Science, or in room 840 in the Department of Physics.

What else?

I like to travel and / or take pictures.