Active Learning for Speeding Up Numerical Experiments in Sensitivity Analysis and Simulation Calibration

Active learning is a popular machine learning method that enables an algorithm to iteratively select instances to be included in the training set during model training. This approach is suitable for cases where there are not many labeled instances available, unlabeled instances are abundant, and it is possible to label the unlabeled instances, however it is costly to do so. Two specific problems that can particularly benefit from active learning: sensitivity analysis and simulation calibration. Sensitivity analysis can be used to assess the level of confidence that is associated with the outcomes of a study. Simulation calibration is the problem of fine-tuning free model parameters to match the simulation outcome to the observed realities. In this thesis, our detailed numerical study shows that active learning is an effective method that leads to substantial computation time savings in conducting numerical experiments.