Choose timezone
Your profile timezone:
Experimental Particle and Astro-Particle Physics Seminar
→
Europe/Zurich
Description
Abstract: Interpreting signals recorded by particle detectors often requires solving complex inverse problems. In high-energy physics, this task relies heavily on detailed detector simulations that model particle interactions and detector responses. For decades, these simulations have been used primarily as forward prediction tools, with detector calibration and event reconstruction treated as sequential and often separate optimization problems. In this seminar, I will introduce end-to-end differentiable simulations and discuss how they can provide an alternative to state-of-the-art HEP workflows. In differentiable simulations, gradients can be computed through the full detector simulation, enabling calibration and event reconstruction to be performed simultaneously using gradient-based optimization within a single unified framework. I will illustrate this idea through the development of LUCiD, a differentiable simulation framework for optical particle detectors. The framework achieves performance comparable to established non-differentiable reconstruction algorithms while significantly simplifying the analysis workflow. Using this example as a starting point, I will discuss the broader challenges and prospects for differentiable simulations to become a mainstream tool in HEP.