Greedy sensor selection for non-linear models

Sensor networks are used to gather information about the environment and to communicate this to the outside world. Sensor selection is an important design problem as the number of sensors is often limited by resource or economical constraints. In this work, the sensor selection problem for non-linear measurement models in additive Gaussian noise is considered. For this purpose, a greedy algorithm based on two submodular cost functions, namely the weighted frame potential and the weighted log-det, is developed. The proposed greedy algorithm is computationally attractive as compared to existing sensor selection solvers for non-linear models. The submodular cost ensures near-optimality of the greedy algorithm.