Catalog-free modeling of galaxy types in deep images

Massive dimensional reduction with neural networks

Institut d’Astrophysique de Paris, Sorbonne Université, CNRS, UMR 7095, 98 bis boulevard Arago, 75014 Paris, France
e-mail: florian.livet@iap.fr, livetflorian@gmail.com

Received: 20 January 2021
Accepted: 5 May 2021

Abstract

Context. Current models of galaxy evolution are constrained by the analysis of catalogs containing the flux and size of galaxies extracted from multiband deep fields. However, these catalogs contain inevitable observational and extraction-related biases that can be highly correlated. In practice, taking all of these effects simultaneously into account is difficult, and therefore the derived models are inevitably biased as well.

Aims. To address this issue, we use robust likelihood-free methods to infer luminosity function parameters, which is made possible by the massive compression of multiband images using artificial neural networks. This technique makes the use of catalogs unnecessary when observed and simulated multiband deep fields are compared and model parameters are constrained. Because of the efficient data compression, the method is not affected by the required binning of the observables inherent to the use of catalogs.

Methods. A forward-modeling approach generates galaxies of multiple types depending on luminosity function parameters rendered on photometric multiband deep fields that include instrumental and observational characteristics. The simulated and the observed images present the same selection effects and can therefore be properly compared. We trained a fully convolutional neural network to extract the most model-parameter-sensitive summary statistics out of these realistic simulations, shrinking the dimensionality of the summary space to the number of parameters in the model. Finally, using the trained network to compress both observed and simulated deep fields, the model parameter values were constrained through population Monte Carlo likelihood-free inference.

Results. Using synthetic photometric multiband deep fields similar to previously reported CFHTLS and WIRDS D1/D2 deep fields and massively compressing them through the convolutional neural network, we demonstrate the robustness, accuracy, and consistency of this new catalog-free inference method. We are able to constrain the parameters of luminosity functions of different types of galaxies, and our results are fully compatible with the classic catalog-extraction approaches.