Enhanced Air Quality Inference via Multi-View Learning With Mobile Sensing Memory

Fine-grained air quality can provide essential urban environmental information for administrators and residents. With advances in communication sensing technologies, low-cost portable sensors installed on vehicles enable high-coverage monitoring. However, data collected by mobile may be inaccurate inconsistent complex operation environments, which brings the issue of uncertainty. Moreover, due to uncontrolled human activities, coverage nodes is dynamic over time, leading uneven or sparse spatial distribution. To address these challenges, we propose AQI-M 3 , a novel framework fine-grained a ir xmlns:xlink="http://www.w3.org/1999/xlink">q uality xmlns:xlink="http://www.w3.org/1999/xlink">i nference via xmlns:xlink="http://www.w3.org/1999/xlink">m ulti-view learning with obile emory. Specifically, an encoder-decoder structure applied region view modeling dependencies pollution maps. More importantly, gradients are extracted trajectory utilization uncertain data. In addition, memory network designed capture patterns from historical global as complemental guide overcome sampling. Extensive experiments conducted three real-world deployments hybrid systems both static sensors. Experimental results show that our proposed approach outperforms competitive baselines 17% $\sim 29$ % reduction mean absolute error. Furthermore, detailed evaluations demonstrate effectiveness robustness under coverage.