Energy efficient learning and classification for distributed sensing

Abstract

In order to reduce total energy communicated by distributed sensors, we address the problem where the recipient wants to perform supervised learning and classification on the data received from the sensors. Restricting our attention to a noiseless communication setting under simplistic Gaussian source assumptions, we provide algorithms for performing this supervised learning under energy limitations. The key idea we bring in is to approximate the problem of minimizing classification error-probability by minimizing the distortion in recovering the decision variable. Constraings on energy consumption in sensors is brought in by using simplistic circuit models inspired from the Analog-to-Digital Converter (ADCs) models. We provide an algorithm for learning the distribution-parameters of sensor-data under these energy constraints to arrive at high-reliability energy-allocation strategies, while enabling the energy-allocation algorithm to backtrack when the underlying distributions change, or when there is noise in sensed data that can push the algorithm towards a local minimum. Finally, we present numerical results on simulated data demonstrating the promise of the proposed techniques in reducing energy consumption.