Translator Disclaimer
6 July 2018 StarNet: a deep learning analysis of infrared stellar spectra
Author Affiliations +
In an era when spectroscopic surveys are capable of collecting spectra for hundreds of thousands of stars, fast and efficient analysis methods are required to maximize scientific impact. These surveys provide a homogeneous database of stellar spectra that are ideal for machine learning applications. One such application, StarNet, is a convolutional neural network developed to derive stellar labels (temperature, surface gravity, and metallicity) from both SDSS-III APOGEE and synthetic stellar spectra. It has demonstrated excellent precision and accuracy over a wide range of signal-to-noise ratios, when trained on either observed or synthetic spectra. Though StarNet was developed using the high-resolution (R ∼ 20, 000) APOGEE spectra and corresponding ASSeT synthetic grid, we suggest that this technique is applicable to other spectral resolutions, spectral surveys, and wavelength regimes. As a demonstration, we present a version of StarNet trained on lower resolution, R=6000, ASSeT synthetic spectra. This resolution was selected to prepare for spectra delivered by Gemini/NIFS and the forthcoming Gemini/GIRMOS instruments. Results suggest that the stellar parameters determined from this medium-resolution StarNet version can be comparable in precision to the high-resolution APOGEE results. This success can be attributed to (1) a large training set of synthetic spectra (N∼200,000) with a priori stellar labels, and (2) the use of the entire spectrum in the solution rather than a few weighted windows, which is common in other automated spectral analysis methods (e.g. FERRE). Remaining challenges in our StarNet applications include rectification, continuum normalization, and wavelength coverage. Here with present preliminary results on the impact of imperfect continuum rectification when compared to normalized synthetic data. Solutions to these problems will contribute to efficient spectroscopic surveys, their data reduction pipelines, and the precision in their post-data products (for the planned Maunakea Spectroscopic Explorer).
© (2018) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Collin L. Kielty, Spencer Bialek, S. Fabbro, K. A. Venn, T. O'Briain, F. Jahandar, and S. Monty "StarNet: a deep learning analysis of infrared stellar spectra", Proc. SPIE 10707, Software and Cyberinfrastructure for Astronomy V, 107072W (6 July 2018);


Back to Top