Star Clusters on FIRE: Classifying Star Clusters in Synthetic Galaxy Images using Machine Learning

Hydrodynamical simulations offer a way to investigate properties of star clusters and classify them according to their shape in a controlled, known environment. In this work, I present a machine learning (ML) algorithm applied to synthetic images of Milky Way-mass galaxies generated from the Latte suite of FIRE-2 simulations. Latte is the first set of galaxy simulations to resolve open star clusters in a fully cosmological context, with cluster masses resolved down to 10^4.6 M_⊙. The superb (2 pc) spatial resolution within the Latte disks provides new opportunities to explore how star clusters can be identified; in this study, I compare manual and ML approaches of identifying star clusters in these simulations. Recent advances in synthetic imaging allow for analysis of mock Hubble Space Telescope (HST) style images, produced in a large range of optical filters spanning from infrared to near-ultraviolet. Here, I present first results comparing the classifications given by humans with the performance of a convolutional neural network (CNN) developed for this project leveraging a wide range of filters. I discuss how the spatial resolution of the simulations affects the performance of the ML algorithm and next steps for improving the performance of the ML algorithm to increase the accuracy of our classifications, relative to human performance. This is a proof of concept study showing our approach can be used on synthetic images at a similar success rate as the recent analysis from the LEGUS and PHANGS-HST collaborations.