Gaussian processes (GPs) are generally regarded as the gold standard surrogate model for emulating computationally expensive computer-based simulators. However, the problem of training GPs as accurately as possible with a minimum number of model evaluations remains a challenging task. We address this problem by suggesting a novel adaptive sampling criterion called VIGF (variance of improvement for global fit). It is the variance of an improvement function which is defined at any location as the square of the difference between the fitted GP emulator and the model output at the nearest site in the current design. At each iteration of the proposed algorithm, a new run is performed where the VIGF criterion is the largest. Then, the new sample is added to the design and the emulator is updated accordingly. The batch version of VIGF is also proposed which can save the user time when parallel computing is available. The applicability of our method is assessed on a number of test functions and its performance is compared with several sequential sampling strategies. The results suggest that our method has a superior performance in predicting the benchmarking functions in most cases.
翻译:Gausian 进程(GPs)一般被视为模拟计算费用昂贵的计算机模拟器的金标准替代模型,然而,以最低数量的模型评价来尽可能准确地培训GPs的问题仍然是一项艰巨的任务。我们提出一个新的适应性抽样标准,称为VIGF(全球适应性改进的变异性),以解决这一问题。这是在任何地点的改进功能的差异,被界定为安装的GP模拟器与当前设计中最接近地点的模型输出之间的差数的正方形。在拟议的算法的每一次迭代中,都在VIGF标准最大的地方进行新的运行。然后,在设计中添加新的样本,并对模拟器进行相应的更新。还提出了VIGF的批量版本,这样可以节省用户在进行平行计算时的时间。我们的方法的适用性根据若干测试功能进行评估,其性能与若干顺序抽样战略进行比较。结果显示,我们的方法在多数情况下预测基准函数方面表现优劣。