ESPAI

About the ESPAI Project

The ESPAI project leverages advanced machine-learning methods to improve the signal purity of XMM-Newton X-ray observations, focusing specifically on data from the EPIC cameras. Its primary objective is to develop an autonomous framework that can identify and mitigate solar-flare contamination, improving data quality while reducing the amount of scientifically useful data that is discarded during filtering.

To tackle this, ESPAI builds a binary, photon-level classifier that labels each detected event as either “flare” or “signal.” The classifier’s output is then used to isolate high-confidence solar-flare photons, which serve as the seed for a generative AI model that produces a large sample of “pure” solar-flare events. This synthetic but physically grounded dataset is designed to support detailed studies of this key source of background in X-ray observations, while also enabling more targeted mitigation in EPIC data processing.

The project delivers openly accessible tools: curated datasets, generative frameworks and source code repositories, all provided under transparent licensing terms. In this way, ESPAI contributes to reproducible science and accelerates discovery for both the astrophysics and AI communities.

Project outputs

• Synthetic datasets: High-fidelity event sequences generated from optimised Variational Autoencoder (VAE) models. Each sample includes relevant photon properties (energy, position) and metadata, derived from XMM-Newton observations, providing a realistic and reliable resource for astrophysical studies and model training.
• Inference and classification tools: Multi-Layer Perceptron (MLP) classifiers that distinguish solar-flare photons from the others within Bad Time Intervals (BTIs). They are robust to data degradation (e.g. dead columns/rows/pixels, “cheesing holes”) and utilise advanced feature engineering like spatial and spatio-temporal photon density.

Contact & Acknowledgements