Classification with waiting times
Abraham J. Wyner · 2002
In this work we seek extensions of a previous study by Wyner and Ziv (1996) to the classification problem with finite memory. The task, broadly, is to decide if observations X are from a known stochastic source Q or from a stochastic source P/spl ne/Q. We introduce an algorithm based upon a simple theorem that quantifies the relationship between waiting times and the probability laws P and Q. We introduce a simple /spl alpha/-significance test of the hypothesis P=Q against the alternative that P/spl ne/Q. We show that this test is not only more simple, but it is more powerful and much more practical.