Machine learning cosmic inflation

Abstract

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