Abstract. State-of-the-art algorithms for solving hard computational problems often expose many parameters whose settings critically affect empirical performance. Manually exploring the resulting combinatorial space of parameter settings is often tedious and unsatisfactory. Automated approaches for finding good parameter settings are becoming increasingly prominent and have recently lead to substantial improvements in the state of the art for solving a variety of computationally challenging problems. However, running such automated algorithm configuration procedures is typically very costly, involving many thousands of invocations of the algorithm to be configured. Here, we study the extent to which parallel computing can come to the rescue. We compare straightforward parallelization by multiple independent runs with a more sophisticated method of parallelizing the model-based configuration procedure SMAC. Empirical results for configuring the MIP solver CPLEX demonstrate that near-optimal speedups can be obtained with up to 16 parallel workers, and that 64 workers can still accomplish challenging configuration tasks that previously took 2 days in 1–2 hours. Overall, we show that our methods make effective use of large-scale parallel resources and thus substantially expand the practical applicability of algorithm configuration methods. 1

Bayesian optimization (BO) aims to minimize a given blackbox function using a model that is updated whenever new evidence about the function becomes available. Here, we address the problem of BO under partially right-censored response data, where in some evaluations we only obtain a lower bound on the function value. The ability to handle such response data allows us to adaptively censor costly function evaluations in minimization problems where the cost of a function evaluation corresponds to the function value. One important application giving rise to such censored data is the runtime-minimizing variant of the algorithm configuration problem: finding settings of a given parametric algorithm that minimize the runtime required for solving problem instances from a given distribution. We demonstrate that terminating slow algorithm runs prematurely and handling the resulting rightcensored observations can substantially improve the state of the art in model-based algorithm configuration. 1