AN ONLINE STOCHASTIC KERNEL MACHINE FOR ROBUST SIGNAL CLASSIFICATION

We present a novel variation of online kernel machines in which we exploit a consensus based optimization mechanism to guide the evolution of decision functions drawn from a reproducing kernel Hilbert space (RKHS) such that the entire stationary process observed can be efficiently modeled. We derive an efficient classification algorithm based on these principles such that our algorithm reduces to traditional online kernel machines for the special case in which the consensus based optimization mechanism is switched off. We illustrate the inherent label and input noise resistance of our algorithm for the case of online classification; and derive relevant mistake bounds. The resulting algorithm can find numerous applications such as, for example, in Automatic Target Recognition (ATR) by remote sensing platforms wherein the target being classified tends to typically be persistent over the observation interval.