Can deep-learning outperform traditional techniques for denoising asteroseismic spectra?

Peak-bagging and peak-tagging of asteroseismic modes, which involve their detection, indentification and characterization, are crucial for all downstream inferences of stellar properties. Filtering techniques, such as curvelet analysis has been shown to improve signal-to-noise ratio leading to better peak-tagging [1]. In recent times, deep learning techniques have outperformed traditional algorithmic denoising techniques for image denoising for a variety of noise profiles. In this study we explore the applicability of different machine learning frameworks for denoising asteroseismic spectra. [1] Lambert et. al. (2006), A&A, Vol. 454, 3, p1021-1027