Efficient User Incentives in Extreme Edge Sensing: A New Learning-Based Scheme

Abstract

Extreme Edge Sensing (EES) systems utilize the built-in sensors of users' smart devices to collect data from the surrounding environment and employ their processors to carry out edge computing tasks. Emerging as a solution to remote sensing challenges, EES systems are noted for their efficient time and cost management, scalability, and the ability to gather real-time data. Efforts to enhance these systems have focused on improving quality of data (QoD) and coverage, as well as on developing incentive schemes to optimize performance. In this thesis, we assess the impact of users' mobility and availability on the spatiotemporal coverage and QoD of EES systems, considering the heterogeneity of users. We propose a distribution-aware and learning-based dynamic incentive scheme. Specifically, we consider the randomness of users' mobility and velocity using a 2-dimensional random waypoint (RWP) model and support the learning-based incentive scheme with a long short-term memory (LSTM) model. The LSTM model utilizes the users' historical data to predict their availability to perform the sensing tasks. The proposed scheme is used to enhance system performance and effectively manage the trade-off between quality and cost, by recruiting users based on the required quality and cost constraints to meet the minimum quality requirement within a constrained budget. Our findings demonstrate that the proposed scheme enhances the spatiotemporal coverage and QoD in EES systems.