A Method to Distinguish Quiescent and Dusty Star-forming Galaxies with Machine Learning. (arXiv:2002.05729v1 [astro-ph.GA])
<a href="http://arxiv.org/find/astro-ph/1/au:+Steinhardt_C/0/1/0/all/0/1">Charles L. Steinhardt</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Weaver_J/0/1/0/all/0/1">John R. Weaver</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Maxfield_J/0/1/0/all/0/1">Jack Maxfield</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Davidzon_I/0/1/0/all/0/1">Iary Davidzon</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Faisst_A/0/1/0/all/0/1">Andreas L. Faisst</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Masters_D/0/1/0/all/0/1">Dan Masters</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Schemel_M/0/1/0/all/0/1">Madeline Schemel</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Toft_S/0/1/0/all/0/1">Sune Toft</a>

Large photometric surveys provide a rich source of observations of quiescent
galaxies, including a surprisingly large population at z>1. However,
identifying large, but clean, samples of quiescent galaxies has proven
difficult because of their near-degeneracy with interlopers such as dusty,
star-forming galaxies. We describe a new technique for selecting quiescent
galaxies based upon t-distributed stochastic neighbor embedding (t-SNE), an
unsupervised machine learning algorithm for dimensionality reduction. This
t-SNE selection provides an improvement both over UVJ, removing interlopers
which otherwise would pass color selection, and over photometric template
fitting, more strongly towards high redshift. Due to the similarity between the
colors of high- and low-redshift quiescent galaxies, under our assumptions
t-SNE outperforms template fitting in 63% of trials at redshifts where a large
training sample already exists. It also may be able to select quiescent
galaxies more efficiently at higher redshifts than the training sample.

Large photometric surveys provide a rich source of observations of quiescent
galaxies, including a surprisingly large population at z>1. However,
identifying large, but clean, samples of quiescent galaxies has proven
difficult because of their near-degeneracy with interlopers such as dusty,
star-forming galaxies. We describe a new technique for selecting quiescent
galaxies based upon t-distributed stochastic neighbor embedding (t-SNE), an
unsupervised machine learning algorithm for dimensionality reduction. This
t-SNE selection provides an improvement both over UVJ, removing interlopers
which otherwise would pass color selection, and over photometric template
fitting, more strongly towards high redshift. Due to the similarity between the
colors of high- and low-redshift quiescent galaxies, under our assumptions
t-SNE outperforms template fitting in 63% of trials at redshifts where a large
training sample already exists. It also may be able to select quiescent
galaxies more efficiently at higher redshifts than the training sample.

http://arxiv.org/icons/sfx.gif