Adaptive design for Gaussian process regression under censoring
A key objective in engineering problems is to predict an unknown experimental surface over an input domain. In complex physical experiments, this may be hampered by response censoring, which results in a significant loss of information. For such problems, experimental design is paramount for maximizing predictive power using a small number of expensive experimental runs. To tackle this, we propose a novel adaptive design method, called the integrated censored mean-squared error (ICMSE) method. Our ICMSE method first learns the underlying censoring behavior, then adaptively chooses design points which minimize predictive uncertainty under censoring. Under a Gaussian process regression model with product Gaussian correlation function, the proposed ICMSE criterion has a nice closed-form expression, which allows for efficient design optimization. We demonstrate the effectiveness of the ICMSE design in the two real-world applications on surgical planning and wafer manufacturing.
READ FULL TEXT