Due to the computational complexity, only the special case of leave-1-out is widely used [8, 11, 12]. Commonly used non-exhaustive cross validation criteria include m-fold [13, 14], holdout [15], random sampling [16], etc. It is worth mentioning that leave-k-out is the exhaustive ...
In this case, computing the worst case cost implies the solution of a quadratic maximization problem with a computational complexity that is exponential with the prediction horizon. In this paper an efficient upper bound of a quadratic maximization problem over a hypercube is presented. It is based...