Game theoretical signal classification: Application to imperfect or non-cooperative learning

Abstract

This paper studies in theory, and gives a solution to, the following concerns: (1) obtain alternative classification decisions, ranked by some decreasing order of class membership probabilities; (2) imperfect teacher at the learning stage, or effect of labelling errors due to unsupervised learning; (3) non-cooperative teacher manipulating the a-priori class probabilities; (4) unknown a-priori class probabilities. This is carried out by considering a game between the recognition system and the teacher, in a game theoretical framework. Both players will ultimately select “mixed strategies” which are probability distributions over the set of N pattern classes. Within the context of signal classification, these N classes are N alternate signal classes e.g., sources, modelled by autoregressive processes. This approach, and the concerns (1)–(4) are especially relevant for the performance enhancement of a number of acoustic signal classification systems (e.g., seismic exploration, intrusion detection, sonar).