Machine Learning meets the redshift evolution of the CMB Temperature

We present a model independent and non-parametric reconstruction with a Machine Learning algorithm of the redshift evolution of the Cosmic Microwave Background (CMB) temperature from a wide redshift range $z\in \left[0,3\right]$ without assuming any dark energy model, an adiabatic universe or photon number conservation. In particular we use the genetic algorithms which avoid the dependency on an initial prior or a cosmological fiducial model. Through our reconstruction we constrain new physics at late times. We provide novel and updated estimates on the $\beta$ parameter from the parametrisation $\text{T}(z)=\text{T}_0(1+z)^{1-\beta}$, the duality relation $\eta(z)$ and the cosmic opacity parameter $\tau(z)$. Furthermore we place constraints on a spatial varying fine structure constant $\alpha$, which would have signatures in a broad spectrum of physical phenomena such as the CMB anisotropies. Overall we find no evidence of deviations within the $1\sigma$ region from the well established $\Lambda\text{CDM}$ model, thus confirming its predictive potential.