Geospatial Imputation of Urban Mobility Data with Self-Supervised Learning
| dc.contributor.author | Han, Bin | |
| dc.contributor.author | Howe, Bill | |
| dc.date.accessioned | 2023-12-26T18:47:25Z | |
| dc.date.available | 2023-12-26T18:47:25Z | |
| dc.date.issued | 2024-01-03 | |
| dc.identifier.doi | 10.24251/HICSS.2023.677 | |
| dc.identifier.isbn | 978-0-9981331-7-1 | |
| dc.identifier.other | a4e0eb03-2009-463c-abaf-83b0f7c27ff7 | |
| dc.identifier.uri | https://hdl.handle.net/10125/107062 | |
| dc.language.iso | eng | |
| dc.relation.ispartof | Proceedings of the 57th Hawaii International Conference on System Sciences | |
| dc.rights | Attribution-NonCommercial-NoDerivatives 4.0 International | |
| dc.rights.uri | https://creativecommons.org/licenses/by-nc-nd/4.0/ | |
| dc.subject | Geospatial Big Data Analytics | |
| dc.subject | image inpainting | |
| dc.subject | missing data | |
| dc.subject | spatial-temporal | |
| dc.subject | urban computing | |
| dc.title | Geospatial Imputation of Urban Mobility Data with Self-Supervised Learning | |
| dc.type | Conference Paper | |
| dc.type.dcmi | Text | |
| dcterms.abstract | Missing values in urban data can be caused by sensor or software failures, data quality issues, interference from weather events, incomplete data collection, or varying data use regulations; any missing data can render the entire dataset unusable for downstream applications. In our work, we adapt image inpainting techniques to impute large, irregular missing regions in urban settings characterized by temporal dependency and spatial skew. To incorporate temporal information, we adapt computer vision techniques for image inpainting to operate on 3D histograms (2D space + 1D time) commonly used for data exchange in urban settings. To combat spatial skew of urban data --- small dense regions surrounded by large sparse areas, we 1) train simultaneously in space and time, and 2) focus attention on dense regions by biasing the masks used for training to the dense resgions in the data. We evaluate the core model and these two extensions using the NYC taxi data and the NYC bikeshare data, simulating different conditions for missing data. We show that the core model is effective qualitatively and quantitatively, that biased masking during training reduces error, and that the number of timesteps during learning exhibits a tradeoff between model performance and resolution of transient events. | |
| dcterms.extent | 10 pages | |
| prism.startingpage | 5619 |
Files
Original bundle
1 - 1 of 1