Parallel Online Continuous Arcing with a Mixture of Neural Networks

This paper presents a new arcing (boosting) algorithm called POCA, Parallel Online Continuous Arcing. Unlike traditional arcing algorithms (such as Adaboost), which construct an ensemble by adding and training weak learners sequentially on a round-byround basis, training in POCA is performed over an entire ensemble continuously and in parallel. Since members of the ensemble are not frozen after an initial learning period (as in traditional arcing) POCA is able to adapt rapidly to non-stationary environments, and because POCA does not require the explicit storage of exemplar statistics, it is capable of online learning. We present results from experiments conducted using neural network experts which show that POCA is competitive with and more flexible than existing arcing algorithms.