Oral Presentation

A Convolutional Variational Autoencoder for Measuring Galactic Environment: The Star Formation–Density Relation

Author(s): Jun-Sung Moon (Yonsei University), Suk-Jin Yoon (Yonsei University)

Presenter: Jun-Sung Moon (Yonsei University)

Galaxies in denser environments tend to have lower star formation rates. Despite its importance on galaxy evolution, there is no universal definition of "environment." To measure the environment of a given galaxy, here we apply the artificial neural network to the spatial distribution of galaxies in the SDSS. We propose a deep learning model that consists of a convolutional variational autoencoder (VAE) and a supervised classifier. The VAE extracts meaningful features from galaxy distributions, quantifying the environment with a handful of latent parameters. The classifier guides the VAE training and predicts the quenching probability. We find that our model reproduces the observed star formation–density relation well. Not only does the prediction outperform in accuracy any single conventional indicator, but also it successfully captures the effect of anisotropic large-scale structures. We will discuss various applications of the model, such as environmental effect estimation, dimensionality reduction for the local environment, and generation of the fake galaxy distribution.