Galaxy stellar mass estimation with Graph Neural Networks
Estimating the stellar mass of galaxies is essential for comprehending the evolution and dynamics of the Universe. This research focuses on advancing this process implementing an innovative Graph Neural Network (GNN) methodology using observational data from the DESI (Dark Energy Spectroscopic Instrument) Early Data Release, our approach aims to surpass limitations of traditional models like multivariable linear regression and Random Forest. Graph Neural Networks excel in handling irregular and sparse data, making them well-suited for tasks involving the spatial distribution of galaxies. The GNN not only harnesses the intrinsic features of each galaxy, such as fluxes in the R, G, Z, W1, and W2 filters, alongside its redshift Z, but also incorporates the spatial relationship among galaxies through graph construction. Our work represents a pioneering effort, as similar estimations with neural networks have predominantly relied on simulations rather than observational data. Through this research, we aim to uncover intricate complexities within the galactic realm and establishing a new standard for implementing machine learning techniques in observational astrophysics.