Calibration of ML prediction errors for samples in Figs. 5a and c. a) Relative error ϵ_r as a function of the corresponding flux for MIPS1 observations (red points), W4 observations (blue points), and the lower (black points) and upper (gray points) prediction intervals for the ML predictions at 24 μm. b) Distribution of the relative errors in a) (same color code). c) Relative error ϵ_r as a function of the corresponding flux for IRAC4 observations (red points), W3 observations (blue points), and the lower (black points) and upper (gray points) prediction intervals for the ML predictions at 8 μm. d) Distribution of the relative errors in c) (same color code).