A new method for adaptive sequential sampling for learning and parameter estimation

Document Type

Conference Proceeding

Publication Date

7-14-2011

Abstract

Sampling is an important technique for parameter estimation and hypothesis testing widely used in statistical analysis, machine learning and knowledge discovery. In contrast to batch sampling methods in which the number of samples is known in advance, adaptive sequential sampling gets samples one by one in an on-line fashion without a pre-defined sample size. The stopping condition in such adaptive sampling scheme is dynamically determined by the random samples seen so far. In this paper, we present a new adaptive sequential sampling method for estimating the mean of a Bernoulli random variable. We define the termination conditions for controlling the absolute and relative errors. We also briefly present a preliminary theoretical analysis of the proposed sampling method. Empirical simulation results show that our method often uses significantly lower sample size (i.e., the number of sampled instances) while maintaining competitive accuracy and confidence when compared with most existing methods such as that in [14]. Although the theoretical validity of the sampling method is only partially established. we strongly believe that our method should be sound in providing a rigorous guarantee that the estimation results under our scheme have desired accuracy and confidence. © 2011 Springer-Verlag Berlin Heidelberg.

Publication Source (Journal or Book title)

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

First Page

220

Last Page

229

This document is currently not available here.

Share

COinS