Distributed informative sensor determination via sparsity-cognizant matrix decomposition

A novel framework is developed that decomposes a matrix into sparse factors. The sparse matrix decomposition scheme is utilized to determine in a distributed fashion which sensors, in a sensor network, acquire informative data about phenomena of interest. A setting, where the sensor data covariance matrix consists of hidden sparse factors, is considered. The proposed sparsity-cognizant algorithm is used to determine the support of the sparse covariance factors, and subsequently identify the informative sensors. A centralized formulation is given first that relies on norm-one regularization. Then, using the notion of missing covariance entries, we obtain an optimization framework that allows distributed estimation of the unknown sparse factors. The corresponding optimization problems are tackled via simple coordinate descent iterations. Different from existing approaches, the novel utilization of covariance sparsity allows distributed source-informative sensor identification, without the need of knowing the data model parameters.