Trajectory generative models enable mobility data sharing without directly releasing original records, but generated trajectories may retain identifiable mobility patterns and trained models may leak sensitive training information. Existing related reviews mainly focus on trajectory generation methods and privacy protection, while comparisons of privacy attack methods for trajectory generation models remain relatively limited. This survey organizes privacy attacks by attacker access and attack objective, distinguishing black-box attacks using released trajectories or generation interfaces from white-box attacks using model-internal information. We review membership inference, re-identification, reconstruction and training-data extraction, and attribute and property inference. We further examine differences in attack signals and evaluation metrics, while distinguishing direct evidence on trajectory generators from findings in adjacent trajectory tasks and other data domains. Our analysis shows how attacker access, evaluation metrics, and experimental settings affect the interpretation of privacy risks. Finally, we discuss directions for more comparable privacy evaluation and for assessing attacks on diffusion-based trajectory generators and those assisted by large language models.
Urban areas often combine multiple functions and density patterns, making single-label zoning inadequate for inference from mobile phone activity. We examine how distance selection affects multi-label k-nearest-neighbour (k-NN) classification of urban function and density on a regular grid using Telecom Italia Big Data Challenge data for Milan. Eleven distances are evaluated across four multi-label k-NN classifiers and three activity views, incoming calls, outgoing calls, and internet activity, and contextualised with four fixed non-k-NN baselines. Each cell has a 168-dimensional weekly signature and five nonexclusive labels: home, work, high density, medium density, and low density. Performance is assessed with nine metrics under iterative stratified 10-fold cross-validation and Friedman–Nemenyi tests. Canberra attains the strongest overall ranks within the k-NN pipelines, whereas complexity-invariant dynamic time warping performs poorly despite its higher cost. Under the outgoing-call view, the local-k classifier obtains the best Ranking Loss (0.095) and Accuracy (0.742) with Canberra. The results show small but systematic rank differences: Canberra is the strongest candidate for this Milan benchmark, whereas elastic warping is consistently unhelpful under the tested conditions. The Milan case provides a basis for testing these findings in other regions with locally observed mobile activity data and reference labels.
The integration of 3D models and geographic information systems (GIS) is a fundamental yet challenging task for City Information Modeling (CIM), because many 3D models lack valid spatial references for being placed in GIS. Existing studies have attempted to perform geo-referencing for Building Information Models (BIMs) by using building boundaries. Feature points, another critical ingredient of both 3D models and GIS, provide a promising basis for matching 3D model views with remote sensing images. However, feature-point matching over large-scale GIS scenes consumes substantial memory and computational time. To address this issue, this study presents Disgeo, a distributed geo-referencing framework that decomposes large-scale GIS scenes into smaller remote sensing image windows and assigns them to independent geo-referencing units. A feature-point-based distributed geo-referencing algorithm, Disgeor, is then developed by embedding SIFT and ASLFeat into this framework. In this study, the distributed framework was implemented and evaluated on a single workstation with 16 parallel processes rather than on a multi-computer cluster. The results show that the proposed scheme achieved decimeter-level accuracy in the tested cases and reduced memory usage by 24
The K Nearest Neighbors Query (KNNQ) is a fundamental distance-based query with applications in Geographic Information Systems (GIS) and many other domains. The K Nearest Neighbors Join Query (KNNJQ) combines KNNQ with the join operation to find the K nearest neighbors from a dataset P to every object in another dataset Q. Due to the large number of distance computations involved, KNNJQ is an expensive operation that poses a challenge for in-memory systems. In this paper, we propose and evaluate novel in-memory algorithms for KNNQ and KNNJQ using the compact data structure k^2 -tree. The proposed algorithms exploit hierarchical spatial decomposition together with grouping, pre-processing and progressive pruning techniques. In addition, we develop two new plane-sweep algorithms for KNNQ and KNNJQ, that improve the brute-force algorithms by reducing redundant computations. We conduct an extensive experimental evaluation using real-world spatial datasets to evaluate the performance of our KNNQ and KNNJQ algorithms, comparing them with KD-tree and R*-tree based algorithms. Our results highlight the efficiency of pruning techniques in k^2 -trees, which reduce distance computations and response time by up to an order of magnitude in KNNJQ, in comparison with the equivalent KD-tree, R*-tree and plane-sweep algorithms.
Data-driven prediction and decision-making increasingly depend on large-scale data lakes and machine learning (ML) and deep learning (DL) analytics across a wide range of application domains. Nevertheless, although geospatial vector data constitutes a substantial fraction of contemporary datasets, it has remained primarily managed within traditional GIS and database systems and has benefited only to a limited extent from recent advances in artificial intelligence (AI). This limitation has been largely attributable to the intrinsic characteristics of geospatial information, in particular when represented in vector format, like multidimensional spatial coordinates, scale and projection effects, variable-size geometries, and a broad spectrum of spatial operations, which are difficult to reconcile with the fixed-dimensional inputs typically required by DL models. This paper advanced DL-enabled geospatial analytics by enabling AI models and agents to provide native access, interpretation, and query processing over geospatial vector datasets. Rather than proposing novel architectures, we introduced a methodology for leveraging existing DL models by properly encoding both geospatial inputs and the spatial operations to be estimated. We investigated three representation/architecture families: (i) image-like, georeferenced histograms processed with ResNet and UNet, (ii) point- and graph-based variants of PointNet++, and (iii) fixed-length geometry embeddings obtained by extending Poly2Vec combined with a Transformer model. We instantiated these approaches on three representative geospatial operators, spatial synopsis, spatial clustering, and walkability estimation, and evaluated them using synthetic and real-world datasets. The results demonstrated that all three families could achieve high accuracy in specific configurations with distinct trade-offs: graph-based models were most effective for dense data and short-range interactions, whereas image- and vector-based approaches exhibited superior scalability due to fixed-size inputs, and the image-based approach was particularly effective for multi-dataset operations.
