Scalable learning and knowledge discovery via adaptive sampling

Scalability is an important issue in data mining and knowledge discovery in real-world applications where hugedata sets often render ordinary learning algorithms infeasible. As an important technique for parameter estimation and hypothesis testing widely used in statistical analysis, random sampling can be exploited to address the issue of scalable learning and knowledge discovery. Adaptive sampling is typically more efficient than traditional batch sampling methods because it can determine the sample size based on the samples seen so far. Recently a new adaptive sampling method for estimating the mean of a Bernoulli random variable was proposed in [2], which was empirically shown to require significantly lower sample size (i.e., the number of sampled instances) while maintaining competitive accuracy and confidence when compared with existing approaches. This paper presents theoretical analysis of properties of the proposed sampling method, as well as a brief outline on how to utilize the proposed sampling method to develop an efficient ensemble learning method with Boosting.