The largest scale perturbations: A window on the physics of the beginning

I present results of new statistical techniques for the interpretation of the temperature and polarization maps and power spectra of the cosmic microwave background. We show that the power deficit at low ℓ in the WMAP1 data is consistent with a statistical fluctuation at the 10% level; that future high S/N maps of the temperature and polarization anisotropies can be combined into a reconstruction of the metric perturbations imprinted during inflation; and that machine learning techniques can accelerate cosmological parameter estimation by orders of magnitude while being highly accurate and robust.