Facial beauty and statistical inference

Facial attractiveness is a paradigm of a complex phenomenon depending on many variables through complex criteria, difficult to infer. We present a novel experimental method allowing for a probabilistic description of the single subject preferred modifications of a reference facial image. Our method allows to unveil the essential subjective character of facial attractiveness. Furthermore, we will compare various methods of unsupervised inference (mainly Maximum Entropy and Restricted Boltzmann Machines) of the database of facial modifications according to various subjects, described in terms of Cartesian landmark coordinates of the facial images.