Decoupling the unfairness propagation chain in crowd sensing and learning systems for spatio-temporal urban monitoring

Abstract

In smart cities, urban monitoring systems rely on advanced mobile sensing and learning technologies to track large-scale urban systems and provide efficient urban services in real time. However, the fidelity and amount of sensors deployed at different geo-communities are closely related to their socioeconomic conditions, demographics, and entrenched geographic patterns, causing inequal sensing opportunities across communities. The biased sensing data contain distorted spatio-temporal patterns of undersensed community, inducing unfairness in subsequent algorithmic prediction and decision-making. This work characterize this unfairness propagation chain of sensing - learning - decision-making process. We introduce the first formal mathematical definitions to quantify and decouple community-level unfairness induced by joint cascading effects of sensing inequality and algorithmic bias. Our real-world experiments with vehicular crowdsensing system in Cangzhou, China verifies that sensing inequality, especially community-level gap of sensor fidelity, result in large fairness gap in spatio-temporal data imputation task. Our preliminary results show that sensing inequality amplifies the algorithmic bias. This work is a critical first step in formally defining and understanding unfairness propagation in intelligent spatio-temporal urban monitoring system.