Distributed kernel least squares for nonlinear regression applied to sensor networks

In this paper, we address the task of distributed nonlinear regression. For this, we exploit kernel methods which can cope with nonlinear regression tasks and a consensus-based approach to derive a distributed scheme. Both techniques are combined and a distributed kernel-based least squares algorithm for nonlinear function regression is proposed. We apply our algorithm to sensor networks and the distributed estimation of diffusion fields which are known to be highly nonlinear. Performance evaluations regarding static and time-varying fields with multiple sources and arbitrary network topologies are provided showing a successful reconstruction. For the tracking of time-varying fields our proposed algorithm outperforms the state of the art.