Sequential Classification with Empirically Observed Statistics

Abstract

Motivated by real-world machine learning applications, we consider the binary statistical classification task in the sequential setting where the generating distributions are unknown and only empirically sampled sequences are available to the decision maker. Then, the decision maker is tasked to classify a test sequence which is known to be generated according to either one of two distributions. The decision maker wishes to perform the classification task with minimum number of the test samples, so, at each step, it declares either “1”, “2” or “give me one more test sample”. We propose a classifier and analyze the type-I and type-II error probabilities. Also, we show the advantage of our sequential scheme compared to the existing non-sequential classifiers.