A Fast Algorithm of Sensor Selection for Non-linear Models

Sensor selection involves determining the optimal subset of sensors to minimize uncertainty in estimating the unknown parameter. In this paper, we assume that the uncertainty is determined by the mean square error (MSE) and select the frame potential (FP) as the proxy for MSE. We introduce a fast algorithm, named Fast Frank-Wolfe Algorithm for non-linear models (FFWNL), to address the sensor selection problems for nonlinear models. This algorithm based on the multilinear extension of the weighted FP and is the fastest algorithm up to now when the dimension of the unknown parameter is low. Additionally, we provide performance bounds for FFWNL in a special case and validate its advantages through numerical experiments.