Neural network-based preprocessing to estimate the parameters of the X-ray emission of a single-temperature thermal plasma

Abstract

We present data preprocessing based on an artificial neural network to estimate the parameters of the X-ray emission spectra of a single-temperature thermal plasma. The method finds appropriate parameters close to the global optimum. The neural network is designed to learn the parameters of the thermal plasma (temperature, abundance, normalisation, and redshift) of the input spectra. After training using 9000 simulated X-ray spectra, the network has grown to predict all the unknown parameters with uncertainties of about a few percent. The performance dependence on the network structure has been studied. We applied the neural network to an actual high-resolution spectrum obtained with Hitomi. The predicted plasma parameters agreed with the known best-fit parameters of the Perseus cluster within 10% uncertainties. The result shows a possibility that neural networks trained by simulated data can be useful to extract a feature built in the data, which would reduce human-intensive preprocessing costs before detailed spectral analysis, and help us make the best use of large quantities of spectral data coming in the next decades.

Neural network-based anomaly detection for high-resolution X-ray spectroscopy

“…The anomaly detection also benefits the data preprocessing of X-ray spectroscopy, which we proposed in Ichinohe et al (2018). Assuming a certain spectral model, this method enables us to know the initial parameters for the spectral fitting at an accuracy of a few percent.…”

“…We prepared the training dataset using X-ray spectral simulation following the same procedure described in Ichinohe et al (2018). We used the fakeit command in XSPEC (version 12.10.0c) and Hitomi response files to generate simulated spectra.…”

“…Recently, Ichinohe et al (2018) demonstrated that the technique of deep learning can actually be applied to the parameter estimation problem of X-ray spectroscopy, under the assumption that the model is known and simple (i.e. single-temperature plasma in collisional ionisation equilibrium).…”

High-Throughput Phenotyping Approach for Screening Major Carotenoids of Tomato by Handheld Raman Spectroscopy Using Chemometric Methods

“…For optimization, adaptive moment estimation (ADAM) was used with default parameters recommended by Kingma and Ba [ 37 ]. The neurons in the hidden layer were activated using rectified linear unit (ReLU) activation function, while Softmax function was used for the neurons of the output layer to compute the probability of classes [ 38 ]. Loss was calculated with categorical cross entropy (CCE) as an error function to measure the network performance [ 31 ].…”

X-ray study of spatial structures in Tycho’s supernova remnant using unsupervised deep learning

“…Recent computational advances have made it possible to train DNNs at a reasonable time and cost, and such techniques have become very popular in many areas, including analyses of astrophysical images (Hezaveh et al 2017;Kimura et al 2017), spectra (Ichinohe et al 2018;Ichinohe & Yamada 2019), light curves (Charnock & Moss 2017;Shallue & Vanderburg 2018), and telescope events (Shilon et al 2018).…”