Cycle-StarNet: Bridging the Gap between Theory and Data by Leveraging Large Data Sets

Abstract

Advancements in stellar spectroscopy data acquisition have made it necessary to accomplish similar improvements in efficient data analysis techniques. Current automated methods for analyzing spectra are either (a) data driven, which requires prior knowledge of stellar parameters and elemental abundances, or (b) based on theoretical synthetic models that are susceptible to the gap between theory and practice. In this study, we present a hybrid generative domain-adaptation method that turns simulated stellar spectra into realistic spectra by applying unsupervised learning to large spectroscopic surveys. We apply our technique to the APOGEE H-band spectra at R = 22,500 and the Kurucz synthetic models. As a proof of concept, two case studies are presented. The first is the calibration of synthetic data to become consistent with observations. To accomplish this, synthetic models are morphed into spectra that resemble observations, thereby reducing the gap between theory and observations. Fitting the observed spectra shows an improved average reduced from 1.97 to 1.22, along with a mean residual reduced from 0.16 to −0.01 in normalized flux. The second case study is the identification of the elemental source of missing spectral lines in the synthetic modeling. A mock data set is used to show that absorption lines can be recovered when they are absent in one of the domains. This method can be applied to other fields that use large data sets and are currently limited by modeling accuracy. The code used in this study is made publicly available on GitHub (https://github.com/teaghan/Cycle_SN).

Zeta-Payne: A Fully Automated Spectrum Analysis Algorithm for the Milky Way Mapper Program of the SDSS-V Survey

“…When teamed up with an optimization algorithm, the approach offers a fast way to analyze large volumes of spectroscopic data in the parameter space the neural network has been trained in. Even a higher level of data analysis is achieved with ML-based algorithms that employ the method of domain adaptation and offer a unique opportunity to improve theoretical models by learning from actual observations, as for example realized in the CYCLE-STARNET algorithm (O'Briain et al 2021).…”

The Spectroscopic Binaries from the LAMOST Medium-resolution Survey. I. Searching for Double-lined Spectroscopic Binaries with a Convolutional Neural Network

“…Recently, this technique and its variants, such as a CNN, have become more and more popular in stellar spectroscopy for high computing performance. They have been extensively used in forward modeling (Ting et al 2019;Xiang et al 2019), backward modeling (Leung & Bovy 2019;Guiglion et al 2020;Wang et al 2020), and generative modeling (Yang et al 2021b;O'Briain et al 2021) of stellar spectra. In this work, we use a CNN discriminative model to cope with the binary classification problem.…”

Unsupervised Learning for Stellar Spectra with Deep Normalizing Flows

“…But more generally, the statistical description of the distribution of spectra through Mendis has a wide range of applications beyond the few case studies explored here. For example, Mendis can serve as a bridge for supervised and unsupervised learning -the learned distribution can serve as the prior distribution for domain adaption to close the model-data synthetic gap (O'Briain et al, 2021). Additionally, the distribution can serve as the basis for semi-supervision and few-shots learning with a limited number of stars with high-fidelity labels.…”