This submission was thoroughly reviewed and discussed among the reviewers. The conclusion is that the paper contains some interesting ideas and overall deserves acceptance but this was not an unanimous case as the theory part is thin and the justification of 'better performance' is moderate. Please further justify the latter, e.g. by comparing numerically the obtained overlap not only with the classical BH but also with the optimal overlap (see papers of Mossel-Neeman-Sly 13, Deshpande-Abbe-Monatanari 15 and Coja-Oghlan-Krzakala-Perkins-Zdeborova 16).