AstroFIt 2 – COFUND fellow since September 1, 2017
INAF Research Centre: Osservatorio Astronomico di Padova
Email: mario.pasquato at inaf.it
- Finding Black Holes with Artificial Intelligence
- Finding IMBHs in Star Clusters with Machine Learning
- Seeing in the dark: spotting dark remnant subsystems and intermediate mass black holes in star clusters with machine learning
- Clustering Clusters: Unsupervised machine learning on globular cluster structural parameters
- Weighing the IMBH candidate CO-0.40-0.22∗ in the Galactic Centre (MNRAS, 17/08/2018)
- Blue straggler bimodality: a Brownian motion model (ApJ, 29/10/2018)
Project title: ARTIStIC – ARTificial Intelligence Search for Intermediate-mass black holes in star Clusters
The recent detection of Gravitational Waves (GW) emitted by a Black-Hole (BH) merger resulting in a ~60 MSun BH rekindled interest in Intermediate Mass BHs (IMBHs; in the 50 – 105 MSun range). The dynamical IMBH formation scenarios suggest IMBHs are likely found in dense star clusters. Besides being GW sources, IMBHs are crucial in cosmology for facilitating the assembly of supermassive BHs.
However, until now, both direct and indirect IMBH searches in star clusters proved inconclusive. My project introduces a new method for IMBH detection based on Machine Learning (ML) applied to photometric and spectroscopic data of globular and young massive star clusters. The long-term goal of the project is to understand whether star clusters in the local Universe host IMBHs, and to count and characterize hosts. The project takes place in three phases: initially I will run a large set of realistic cluster simulations and produce mock observations of projected density and kinematics profiles. I will then use ML algorithms to classify these simulated datasets into IMBH-host versus non-hosts. Finally, I will apply the models produced by the algorithms to actual observational data, including Gaia kinematics. The end result of my project will be a large set of globular and young clusters classified into IMBH-host/non-host, with an estimate of the confidence in the classification. Moreover, I will release my codebase (ML model training scripts and possibly the trained models themselves) to the astronomical community. My approach is radically different from previous attempts at IMBH detection because it is data driven, i.e. it does not look for a specific signature of IMBH presence, but rather attempts to optimize the usage of all available data to single out IMBH hosts via ML classification algorithms.