Deep Generative Models for Simulation of Energetic Particles through Dense Materials
Contact person: Eirik Gramstad
Keywords: Generative AI, Neural networks, Physics-informed modelling, Subatomic physics
Research groups: High Energy Physics
Department of Physics
Most analyses of particle physics data rely on simulated collision events, where a digital twin of the detection equipment simulates the passage of high energy particles through its material and active layers. This produces a simulacrum of the real data, allowing analysts to test whether they have a full understanding of the detection equipment. Traditionally this has involved detailed calculations of the passage of particles through the material using tools such as the Geant4 toolkit. This is computationally intensive, so recently generative AI tools, which are trained on the full digital twin simulations, have been introduced. This is a very recent innovation and there are major opportunities for investigating new techniques and methods for generative simulation of particle showers in dense materials. Another important aspect when implementing generative models is to properly define what is the "best model". This is not necessarily a trivial task given the stochastic nature of the energy deposits from a particle passing through material.
Applications stretch well beyond particle physics and can include any field where the passage of energetic particles through dense materials has to be simulated.
Methodological research:
- Algorithms of generative AI
- Detailed simulation of subatomic particle showers
- Fast, accurate and efficient simulations methods
Relevant topics from natural sciences or technology:
- High energy particle, nuclear and accelerator physics
- Medical physics
- Space and Radiation Science
External partners:
- EP-SFT (CERN)
- ATLAS Experiment
Mentoring and internship will be offered by a relevant external partner.
