The problem of large search space in stochastic optimization

We have demonstrated that the problem of large search space in stochastic optimization can be effectively attacked. A great deal of computational burden can be shared if we are simulating a set of parametrically different but structurally similar systems. The metaphor here is the data compression scheme used in the transmission of moving pictures; one transmits the first frame followed by the difference in succeeding frame to minimize transmission channel capacity requirements. Similarly, the evaluation of the performances of a set of experiments involve a great deal of commonality that can be shared and leveraged for maximal efficiency. In particular, the computation is well adapted to SIMD massively parallel computers. Also, softening the strict requirement of optimality can often transform an infeasible problem (in terms of computation burden) into a tractable one.<<ETX>>