MILCANN : A neural network assessed tSZ map for galaxy cluster detection

We present the first combination of thermal Sunyaev-Zel'dovich (tSZ) map with a multi-frequency quality assessment of the sky pixels based on Artificial Neural Networks (ANN) aiming at detecting tSZ sources from sub-millimeter observations of the sky by Planck. We construct an adapted full-sky ANN assessment on the fullsky and we present the construction of the resulting filtered and cleaned tSZ map, MILCANN. We show that this combination allows to significantly reduce the noise fluctuations and foreground residuals compared to standard tSZ maps. From the MILCANN map, we constructed the HAD tSZ source catalog that consists of 3969 sources with a purity of 90\%. Finally, We compare this catalog with ancillary catalogs and show that the galaxy-cluster candidates in the HAD catalog are essentially low-mass (down to $M_{500} = 10^{14}$ M$_\odot$) high-redshift (up to $z \leq 1$) galaxy cluster candidates.