DeepSZ: identification of Sunyaev–Zel’dovich galaxy clusters using deep learning

Galaxy clusters identified from the Sunyaev Zel'dovich (SZ) effect are a key ingredient in multi-wavelength cluster-based cosmology. We present comparison between two methods of cluster identification: standard Matched Filter (MF) method SZ finding and using Convolutional Neural Networks (CNN). further implement show results for `combined' identifier. apply to simulated millimeter maps several observing frequencies an SPT-3G-like survey. There some differences methods. The MF requires image pre-processing remove point sources model noise, while CNN very little images. Additionally, tuning hyperparameters takes as input, cutout images sky. Specifically, we use classify whether or not 8 arcmin $\times$ sky contains cluster. compare purity completeness. signal-to-noise ratio depends on both mass redshift. Our CNN, trained given threshold, captures different set than MF, which have SNR below detection threshold. However, tends mis-classify cutouts whose located near edge cutout, can be mitigated with staggered cutouts. leverage complementarity methods, combining scores each identification. completeness alone 0.61, assuming 0.59 0.61. combined classification yields 0.60 0.77, significant increase modest decrease purity. advocate that confidence many lower clusters.