Optimization of computationally expensive simulations with Gaussian processes and parameter uncertainty: Application to cardiovascular surgery

In many applications of simulation-based optimization, the random output variable whose expectation is being optimized is a deterministic function of a low-dimensional random vector. This deterministic function is often expensive to compute, making simulation-based optimization difficult. Motivated by an application in the design of bypass grafts for cardiovascular surgery with uncertainty about input parameters, we use Bayesian methods to design an algorithm that exploits this random vector's low-dimensionality to improve performance.