We consider, for the first time in the modern setting of high- dimensional statistics, the classic problem of optimizing the objective function in regression. We propose an algorithm to compute this optimal objective function that takes into account the dimensionality of the problem.