关键词：二次取样；贝叶斯模型；随机取样
摘 要：This dissertation addresses some of these issues：We begin with an active learning strategy for spectral clustering when the cost of assessing individual similarities is substantial or prohibitive.Next, we consider active learning in Bayesian models.Our third contribution looks at the e ects of randomized subsampling on Gaussian process models that make predictions about outliers and rare events. Finally, we turn to a theoretical evaluation of randomized subsampling for the purpose of inferring rankings of objects.