ACE: Artificial intelligence for galaxy Cluster dEtection

Pourvu: 

Non
Project ACE aims to implement new techniques from machine learning (ML) and artificial intelligence (AI) to detect galaxy clusters in upcoming astronomical surveys.  Convolutional neural networks offer the possibility to vastly improve cluster detection and the construction of cluster catalogs, improvements that will be critical to reach the full scientific potential of ESA’s Euclid space mission, the Rubin Observatory LSST survey, and the Simons Observatory (SO).  These experiments are dedicated to dark energy research and study of large-scale cosmic structure.  Clusters are one of their primary observational tools.  Our research group is currently involved in the preparation of these three cornerstone cosmological experiments, all three of which will begin operations within five years.
Galaxy clusters are unique tools in cosmology and astrophysics.  Their abundance and its evolution are potentially the most powerful observational probes of dark energy and deviations from standard gravity theory. And study of their galaxy populations over cosmic timescales is the only way to solve the outstanding mystery of galaxy quenching.  Not surprisingly, galaxy cluster catalogs are therefore one of the main scientific products of large astronomical surveys. 
Clusters are, however, complex objects that appear in survey images with varying characteristics, and this makes their detection problematic for standard algorithms.  Such algorithms typically linearly combine measures of only a subset of the simplest expected cluster properties, such as total flux and angular extent.  
Here, ML and AI techniques offer the possibility of qualitative, game-changing gains in performance because they excel in this arena. Neural networks can identify complex object properties, unknown a priori, and find more efficient non-linear combinations with the potential to vastly improve object detection.  
We will construct and implement novel galaxy cluster detection methods based on convolutional neural networks (CNNs).  Our objective is to apply the methods to the upcoming surveys from ESA’s Euclidspace mission, the Rubin Observatory and SO.  One of our specific goals is to improve the efficiency of high-redshift cluster detection by combining the optical/NIR galaxy catalogs from Euclid and Rubin with the trans-millimeter data from SO. This would be a pioneering study and an ideal application of ML and AI techniques.  

Responsable: 

Prof. James Bartlett

Services/Groupes: 

Année: 

2021

Formations: 

Thèse

Email du responsable: