Machine learning cosmic inflation

We present a machine learning approach, based on the genetic algorithms (GA), that can be used to reconstruct inflationary potential directly from cosmological data. create pipeline consisting of GA, primordial code and Boltzmann calculate theoretical predictions, cosmic microwave background (CMB) As proof concept, we apply our methodology Planck CMB data explore functional space single-field potentials in nonparametric, yet analytical way. show algorithm easily improves upon vanilla model quadratic inflation proposes slow-roll better suited data, while confirm robustness Starobinsky (and other small-field models). Moreover, using unbinned perform first concrete application GA by searching for oscillatory features an agnostic way, find very significant improvements best featureless potentials, $\mathrm{\ensuremath{\Delta}}{\ensuremath{\chi}}^{2}<\ensuremath{-}20$. These encouraging preliminary results motivate search resonant power spectrum with multimodal distribution frequencies. stress is modular extended sets scenarios, like multifield or theories higher-order derivatives.