NON-PARAMETRIC PATTERN RECOGNITION PART Ⅱ THE NON-DISJOINT CASE

摘要

In Part I of this paper, a non-parametric discrimination technique was proposed. It was shown that when perfect discrimination was possible, this technique achieved this perfection and in certain, cases achieved it with a finite learning phases In this report, properties of this technique are investigated for the case when perfect discrimination is not possible In particular, it is shown, that even under these conditions asymptotic optimality in the risk sense is attained.