Karlsruhe 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 119: Data, AI, Computing 8 (foundational & transformer models)

T 119.2: Talk

Friday, March 8, 2024, 09:15–09:30, Geb. 30.33: MTI

adaptive generative modeling for High-Granularity Calorimeters — •Lorenzo Valente — Institut für Experimentalphysik, University of Hamburg, Germany

Simulating particle colliders in their entirety presents a substantial computational challenge for researchers. Detector simulations are among the most resource-intensive phases of this process. Deep generative models could be a potential solution since they have already been proven to speed up simulations.

The growing volume of data from upcoming high-energy physics experiments, including higher collider luminosities and highly granular calorimeters, requires the development of artificial intelligence algorithms capable of combining knowledge across different domains. Unfortunately, conventional deep learning algorithms struggle with handling multiple datasets. Research in domain adaptation involves creating methodologies to bridge the divide between datasets, enabling the construction of models that exhibit high performance across diverse domains simultaneously.

In this contribution, we illustrate how a more universal domain adaptation approach, utilizing the transfer learning method, enhances the flexibility of the model. Specifically, we showcase its effectiveness in data generation for different calorimeter geometries.

Keywords: Generative Models; Deep Learning; Domain Adaptation