2025 workshop program
Modeling X-ray photon pile-up with machine learning: A data-driven perspective
The work
Abstract
The dynamical range of detectors flown on-board existing and future X-ray observatories does not cover the whole dynamical range of extrasolar X-ray sources. The analysis of bright X-ray sources is complicated by so-called pile-up, which is caused by the high incident photon flux. The non-linear effect distorts the measured spectrum, resulting in inaccurately fitted model parameters, and can even lead to a complete signal loss in extreme cases. Such data are commonly discarded due to a lack of ability to account for the effects of pile-up and to properly model the spectral distortion of the incident signal. As a result, a large number of archival observations remains under-explored. We deploy machine learning techniques to model the effect from a data-driven viewpoint. The goal is to develop a generic pile-up emulator that helps researchers to address the severity of the pile-up and assist in the analysis of the data, both for existing and future X-ray observatories. I will show a proof-of-concept neural network that is trained on simulations and can reconstruct piled-up eROSITA data. I will talk about the biases and applicability of such an algorithm and will outline approaches to extend the training data to the full Chandra and XMM-Newton archive.