NeurIPS Rebuttal for " Reliable training and estimation of variance networks "