abstract: In 1927 G. Birkhoff conjectured that if a billiard in a strictly convex smooth domain is integrable, the domain has to be an ellipse (or a circle). The conjecture is still wide open, and presents remarkable relations with open questions in inverse spectral theory and spectral rigidity. In the talk we show that a version of Birkhoff's conjecture is true for small perturbations of ellipses of small eccentricity. This is joint work with A. Avila and V. Kaloshin