In convex optimization, relax and round / polishing is a procedure by which you perform a local search after coming up with a relaxation, and round into the actual feasible set (such as integers).