On Myopic Strategies For Resource Constrained Informative Sampling

In this work, we investigate data-driven sensor scheduling policies under resource constraints. We consider a scenario in which data is being generated by n discrete-time dynamical processes. At each instant, m <; n measurements can be collected. Based on an auto-regressive (AR) prediction model, we define two loss functions, namely, the prediction error and the measurement error associated with the missing measurements. The problem of choosing the measurements at each step is formulated as an optimization problem over the two loss functions. For n = 2 and m = 1, we explicitly derive the optimal measurement policy, and present an analytical expression for the decision rule. We extend the technique adopted in the aforementioned problem to derive an algorithmic procedure to address the case when m and n are arbitrary. We present simulations to illustrate the performance of the proposed algorithm, and compare it to the case when the parameters of the AR predictor are updated at each step based on the measurements.