Speaker
Description
In-ice radio detection is a rapidly developing field in which detector design choices made now can have a lasting impact on the sensitivity of EeV neutrino searches. While brute-force simulation campaigns are infeasible for large design parameter spaces, differentiable programming makes it possible to compute gradients of a scientific objective with respect to detector design parameters, enabling efficient gradient-based optimization. In this contribution, we present a fully differentiable end-to-end pipeline for in-ice radio neutrino simulation, detection, and reconstruction, by combining a PyTorch-based reimplementation of core NuRadioMC components, generative machine learning, neural network surrogates, and uncertainty estimation through the Fisher information. This framework enables direct optimization of the science objectives with respect to antenna positions and orientations. We will present end-to-end optimized station designs that improve reconstruction.