Non-negative distributed regression for data inference in wireless sensor networks

Wireless sensor networks are designed to perform on inference the environment that they are sensing. Due to the inherent physical characteristics of systems under investigation, non-negativity is a desired constraint that must be imposed on the system parameters in some real-life phenomena sensing tasks. In this paper, we propose a kernel-based machine learning strategy to deal with regression problems. Multiplicative update rules are derived in this context to ensure the non-negativity constraints to be satisfied. Considering the tight energy and bandwidth resource, a distributed algorithm which requires only communication between neighbors is presented. Synthetic data managed by heat diffusion equations are used to test the algorithms and illustrate their tracking capacity.