Research interests

I find it interesting to combine machine learning models with structure that describes our human knowledge of the underlying laws of nature or the symmetries of a system.

One flavour of this is simulation-based inference. Here we combine simulators written by domain experts, neural surrogate models, and a dash of statistics to get the most precise measurements of parameters out of high-dimensional data. This approach has the potential to speed up scientific progress in areas from particle physics to cosmology, from chemistry to robotics, from neuroscience to sociology.

Another direction I am interested in are latent variable models with structured latent spaces. Lately I have been interested in causality, in particular causal representation learning, and how that can help us solve practical problems in interactive learning.

Selected papers

You can find a more or less complete list of my papers on Google Scholar. Here I want to highlight a few favorites:

Causality and interactive learning

Weakly supervised causal representation learning.
Johann Brehmer, Pim de Haan, Phillip Lippe, and Taco Cohen.
NeurIPS 2022. [arXiv] [bibtex]

Deconfounded imitation learning.
Risto Vuorio, Johann Brehmer, Hanno Ackermann, Daniel Dijkman, Taco Cohen, and Pim de Haan.
NeurIPS workshop on deep reinforcement learning (2022). [arXiv]

Generative models and their applications

Flows for simultaneous manifold learning and density estimation.
Johann Brehmer and Kyle Cranmer.
NeurIPS 2020. [arXiv] [conference] [bibtex] [code]

Instance-adaptive video compression: Improving neural codecs by training on the test set.
Ties van Rozendaal, Johann Brehmer, Yunfan Zhang, Reza Pourreza, and Taco Cohen.
Under review (2022). [arXiv] [bibtex]

Simulation-based inference

The frontier of simulation-based inference.
Kyle Cranmer, Johann Brehmer, and Gilles Louppe.
Proceedings of the National Academy of Science 117 (2020). [arXiv] [journal] [bibtex]

Simulation-based inference methods for particle physics.
Johann Brehmer and Kyle Cranmer.
Book chapter in Artificial Intelligence for High-Energy Physics (World Scientific, 2022). [arXiv] [book] [bibtex]

Mining for Dark Matter Substructure: Inferring subhalo population properties from strong lenses with machine learning.
Johann Brehmer, Siddharth Mishra-Sharma, Joeri Hermans, Gilles Louppe, and Kyle Cranmer.
The Astrophysical Journal 886 (2019). [arXiv] [journal] [bibtex] [code]

MadMiner: Machine learning-based inference for particle physics.
Johann Brehmer, Felix Kling, Irina Espejo, and Kyle Cranmer.
Computing and Software for Big Science 4 (2020). [arXiv] [journal] [bibtex] [code]

Likelihood-free inference with an improved cross-entropy estimator.
Markus Stoye, Johann Brehmer, Gilles Louppe, Juan Pavez, and Kyle Cranmer.
NeurIPS workshop on Machine Learning and the Physical Sciences (2019). [arXiv] [workshop] [bibtex]

Mining gold from implicit models to improve likelihood-free inference.
Johann Brehmer, Gilles Louppe, Juan Pavez, and Kyle Cranmer.
Proceedings of the National Academy of Science 117 (2020). [arXiv] [journal] [bibtex]

Constraining effective field theories with machine learning.
Johann Brehmer, Kyle Cranmer, Gilles Louppe, and Juan Pavez.
Physical Review Letters 121 (2018). [arXiv] [journal] [bibtex] [code]

A guide to constraining effective field theories with machine learning.
Johann Brehmer, Kyle Cranmer, Gilles Louppe, and Juan Pavez.
Physical Review D 98 (2018). [arXiv] [journal] [bibtex] [code]

More deep learning and statistics for physics

Hierarchical clustering in particle physics through reinforcement learning.
Johann Brehmer, Sebastian Macaluso, Duccio Pappadopulo, and Kyle Cranmer.
NeurIPS workshop on Machine Learning and the Physical Sciences (2020). [arXiv] [workshop] [bibtex] [code]

Better Higgs Measurements Through Information Geometry.
Johann Brehmer, Kyle Cranmer, Felix Kling, Tilman Plehn.
Physical Review D 95 (2017). [arXiv] [journal] [bibtex]

Code

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

My GitHub page contains the code used for most of my pre-Qualcomm papers.

Talks

Over the years, I had a lot of fun presenting our research in different places. See this repository for a list and PDF versions of my slides. I'm always happy to talk about our recent papers.

Contact

Write me at jbrehmer@qti.qualcomm.com (work) mail@johannbrehmer.de (personal), connect on Twitter, or come visit me in Amsterdam!

Not work

I like travelling and taking pictures.