Machine learning for X-ray astronomical spectroscopy
Description
We use Simulation Based Inference to construct 34000 artificial spectra that are representative of observed Active Galactic Nuclei X-ray spectra with NASA's NuSTAR X-ray telescope. Based on previous literature results for these spectra, we train, validate, and test a new neural network model, ML-mytorus, which predicts best values for key physical parameters fast, automatically, and with very high degree of accuracy, including error estimates. We make the code publicly available, and set up a dedicated webpage for the community to upload any spectrum and use the NN model. A publication is about to be submitted for review.
Detailed example
Predictions for astrophysical parameters with uncertainties
AI / analytics pattern
Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relationships.
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
Reproducibility: Standard spectral fitting is done interactively with human decision making. Automation and speed: Using the trained neural network model is completely automated, and takes just seconds for a given spectrum (compare to hours/days for interactive fitting). Openly available: Anyone can go to our webpage, load X-ray spectra in 2-column format and obtain results in seconds by clicking on a single button. Extensibility: The methodology can be extended for other X-ray (and non-X-ray) telescopes producing spectra, including for other types of astronomical observations, provided telescope and physical model characteristics are known. Training only needs a modest number (tens...) of existing previous fitted observed spectra, which are statistically perturbed to produce literally thousands of simulated spectra for training. Simulations, training and validation take no more than a few days with modest high-performance computing use.
Controls / human review
ATO: Not reported; PIA: Not published