Social Learning Based Inference for Crowdsensing in Mobile Social Networks

Mobile communication technology provides more service paradigms to social networks, allowing the development of mobile social networks (MSNs). An important scenario of MSNs is crowdsensing, which takes advantage of simple sensing and computation abilities on the portable devices of ordinary people, and fuses the sensing results to accomplish large-scale tasks. In crowdsensing, the integration of individual sensing data from users is of great significance, yet highly depends on the goal of tasks. In this paper, we propose a high-level distributed cooperative environmental state inference scheme based on non-Bayesian social learning, which can be applied to various crowdsensing tasks, e.g., traffic monitoring, air quality monitoring, and weather forecasting. In the proposed scheme, users exchange information with their neighbors and cooperatively infer the hidden state, which is the goal of the crowdsensing task but cannot be measured directly. We prove theoretically that every user is able to asymptotically learn the hidden state, even when the users locations and relationships keep changing, and when some users cannot observe signals or provide their own inferences. We also optimize the weight matrix in the fusion step by maximizing the learning speed.