Every Data Point Count: Multi-Task Data Inference Without Historical Data in Sparse Mobile Crowdsensing
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 23035-23051
DOI: 10.1109/tmc.2026.3725800 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Xiangping Zheng und 7 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Shandong Provincial Natural Science Foundation, Hainan Provincial Natural Science Foundation of China.
Abstract (Originalsprache)
Sparse mobile crowdsensing collects data from sensed areas and infers information for unsensed regions, thereby reducing data collection costs. Previous studies have primarily focused on leveraging complete historical data for sparse single-type data inference. However, real-world scenarios typically involve multiple types of data, and acquiring large amounts of complete historical data is often impractical. To address these challenges, we propose MSTGFI (Multi-Site multi-Task Graph Fusion Inference), a novel framework that builds upon intra-task correlations (correlations among the same type of data across different sites) and inter-task correlations (correlations among different types of data within the same site), while also further exploiting inherent spatial dependencies, thereby fully leveraging spatiotemporal information from limited observations to tackle high sparsity challenges. Specifically, MSTGFI first introduces a multi-site multi-task fusion imputation module that exploits both inter- and intra-task correlations to extract spatiotemporal features from multi-type sparse data, enabling collaborative imputation across heterogeneous data sources. Then, we design a spatial graph masked autoencoder capable of modeling complex spatial dependencies to further enhance reconstruction performance. Finally, we conduct comprehensive qualitative and quantitative experiments on two real-world datasets containing multiple types of sensing data. The results demonstrate that our method significantly outperforms state-of-the-art baselines across various sparsity levels.
Förderung
- National Natural Science Foundation of China
- Shandong Provincial Natural Science Foundation
- Hainan Provincial Natural Science Foundation of China
- Qingdao Municipal Natural Science Foundation
Zitieren
APA
Zheng, X., Wang, Z., Tu, C., Yu, Z., Ren, B., Wang, H., Huang, R., & Yu, Z. (2026). Every Data Point Count: Multi-Task Data Inference Without Historical Data in Sparse Mobile Crowdsensing. IEEE Transactions on Mobile Computing, 25(12), 23035-23051. https://doi.org/10.1109/tmc.2026.3725800
Vancouver
Zheng X, Wang Z, Tu C, Yu Z, Ren B, Wang H, et al. Every Data Point Count: Multi-Task Data Inference Without Historical Data in Sparse Mobile Crowdsensing. IEEE Transactions on Mobile Computing. 2026;25(12):23035-23051. doi:10.1109/tmc.2026.3725800
BibTeX
@article{Zheng2026,
author = {Zheng, Xiangping and Wang, Zhi and Tu, Chunyu and Yu, Zhiyong and Ren, Bin and Wang, Huachuan and Huang, Runhe and Yu, Zhiwen},
title = {Every Data Point Count: Multi-Task Data Inference Without Historical Data in Sparse Mobile Crowdsensing},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {23035--23051},
doi = {10.1109/tmc.2026.3725800}
}
EUVIMED