Oct 18 – 22, 2021
INAF Osservatorio Astrofisico di Arcetri (Italy) and online
Europe/Rome timezone

Reimagining large spectroscopic data with unsupervised machine learning generative models

Oct 21, 2021, 12:20 PM


Yuan-Sen Ting (Australian National University)


Modern-day machine learning generative models allow us to directly model the distribution of the observed spectroscopic data, even when the stellar labels are absent. In recent years, we have seen the explosion of studies in terms of supervised machine learning. However, the exploration of unsupervised generative models in stellar spectroscopy is, unfortunately, lagging behind. In this talk, I will discuss various unsupervised machine learning methods, including generative models and normalizing flows and their pros and cons. I will demonstrate how unsupervised generative models can uncover missing atomic features, auto-calibrate imperfect models, and detect outlier spectra without needing a predefined training set with stellar labels.

Type contributed talk

Primary author

Yuan-Sen Ting (Australian National University)

Presentation materials