Résumé / Abstract Journal-club_Doctorants

Journal-club Doctorants / Journal-club PhD students

« Galaxy evolution modelling with simulated images, Bayesian inference and dimensionality reduction through neural network. »

Florian Livet
Institut d'Astrophysique de Paris (Paris, France)

Nowadays, the study of galaxy evolution is fed with a large set of deep photometric surveys with long exposure time and a wide color range. The classic approach is to extract catalogs of galaxies from these surveys and to model how their luminosities, colors and sizes evolve with redshift. However, the catalogs are often incomplete (Malmquist bias, Eddington bias, confusion, cosmological dimming, stellar contamination, Galactic extinction, K-corrections), therefore the evolution models are often poorly constrained. These spatially-dependent selection effects interact with one another, but it is very hard to express in what extent. The classic approach is then an inverse problem which seeks to model these biases and to correct their effects from the extracted catalogs (Marzke 1998, Taghizadeh-Popp et al. 2015). Therefore, the models derived from these catalogs tend to be biased in a non-trivial way.

In order to avoid solving a hard inverse problem, another reliable technique is the forward-modeling (Carassou et al. 2017). A list of modeled galaxies is passed through a virtual telescope, which is an exact copy of the telescope used for the real survey with the same characteristics and the same observational conditions. This virtual telescope uses the same filters, the same exposure time, thus leads to the same cosmological and instrumental biases described above. This method allows a direct comparison between a real deep survey passed through the real telescope and a simulated deep survey passed through the exact virtual copy of the real telescope. The key-point of this method is that the comparison is done in the observational space and consequently the same selection biases affect both the real and the simulated surveys.

Because of the high dimensionality of the problem, we use a deep learning method to reduce the dimension of the images in order to keep only the essential information: the neural network algorithm developped by Charnock et al. (2018) which seeks to maximize the Fisher information of the images. This allows us to understand the statistical impact of each parameter of our model on the information content of the images, and to optimize the model of galaxy evolution.
vendredi 17 janvier 2020 - 16:00
Salle du Conseil, Institut d'Astrophysique
Page web du journal-club / Journal-club's webpage