Artificial Intelligence as a scientific method has been transforming society and essentially all fields of research for the last one or two decades. Historically, particle physics has been a driver of this development, and by now all kinds of numerical approaches to fundamental physics benefit from a vast landscape of AI-research. This school will train particle physicists to develop machine learning as a transformative addition to our numerical toolbox, and identify new and interesting research questions which we can answer using machine learning.
Particle physics, experiment and theory, are benefiting from modern methods of data science in many ways. Simulations for the LHC and other experiments are asking for ML-applications, to improve the precision and speed of the simulation or make use of inverted simulation chains. A crucial part of this simulation task are precision calculations in quantum field theory, which combine computer-expensive loop calculations with complex phase spaces, and ML-parton densities. Subjet physics has developed into an inspiring topic between theory and experiment, and jet taggers are only the first and obvious ML-application in particle physics. Searches for expected or unexpected new effects are at the heart of LHC physics and will be revolutionized by ML-concepts like model-independent anomaly searches. Forward simulations combined with likelihood-free inference and inverse simulations or unfolding are the analyses methods of the future, at the LHC and elsewhere. Global analyses, for instance combining effective theory frameworks with modern data science tools, are common all over particle physics. Many of these aspect can be directly transferred to cosmology.
Technically, many of these tasks are related to classic problems, like density estimates or incompletely defined inverse problems. Once these tasks grow in complexity, ML-methods are the way to go. Beyond classic problems, studies of the string landscape have been shown to benefit from modern ML-concepts like reinforcement learning. Symmetry-aware neural networks or neural networks learning symmetric representations are a lively new field with a natural link to fundamental physics. Symbolic regression might even allow us to invert the one-way path from formulas to numerics in LHC physics and in cosmology. All of these new methods are being developed right now --- in industry, life sciences, medicine, and physics. For (theoretical) particle physics it is crucial that we learn and further develop these new methods. This school will cover a wide range of topics, from compact introductions to different aspects of modern machine learning to cutting-edge applications in particle physics and neighboring fields.
There are two requirements for all participants: (i) enthusiasm for and a solid background knowledge in particle physics and its numerical methods; (ii) enthusiasm for modern machine learning and basic knowledge of the corresponding Tensorflow and PyTorch tools. This also means are not exclusively looking for theorists, experimental particle physicists with a solid interest in analysis and simulations are very welcome to join us. Please ask your letter writers to comment on these aspects explicitly.
Unlike many other schools, tutorials and code training are not going to be the main focus of this more conceptual program. However, there will be tutorials as part of some lectures. So please bring your laptop and check the program (first lecture by a given lecturer) for computer instructions.
Finally, we are not meeting at the MITP, but in Oppenheim, a 25 minute S-Bahn ride from Mainz main station. Our conference venue is the Altes Amtsgericht, and as all of Oppenheim the hotel Amtsgericht's Blick is in short walking distance from there. The school starts with Dinner at the hotel on Sunday night registration will be at the conference site on Monday morning