Benoit Garbinato

Vaibhav Kulkarni, Natasa Tagasovska, Thibault Vatter et al.

Mobility datasets are fundamental for evaluating algorithms pertaining to geographic information systems and facilitating experimental reproducibility. But privacy implications restrict sharing such datasets, as even aggregated location-data is vulnerable to membership inference attacks. Current synthetic mobility dataset generators attempt to superficially match a priori modeled mobility characteristics which do not accurately reflect the real-world characteristics. Modeling human mobility to generate synthetic yet semantically and statistically realistic trajectories is therefore crucial for publishing trajectory datasets having satisfactory utility level while preserving user privacy. Specifically, long-range dependencies inherent to human mobility are challenging to capture with both discriminative and generative models. In this paper, we benchmark the performance of recurrent neural architectures (RNNs), generative adversarial networks (GANs) and nonparametric copulas to generate synthetic mobility traces. We evaluate the generated trajectories with respect to their geographic and semantic similarity, circadian rhythms, long-range dependencies, training and generation time. We also include two sample tests to assess statistical similarity between the observed and simulated distributions, and we analyze the privacy tradeoffs with respect to membership inference and location-sequence attacks.

Bertil Chapuis, Benoit Garbinato

Finding trajectories and discovering motifs that are similar in large datasets is a central problem for a wide range of applications. Solutions addressing this problem usually rely on spatial indexing and on the computation of a similarity measure in polynomial time. Although effective in the context of sparse trajectory datasets, this approach is too expensive in the context of dense datasets, where many trajectories potentially match with a given query. In this paper, we apply fingerprinting, a copy-detection mechanism used in the context of textual data, to trajectories. To this end, we fingerprint trajectories with geodabs, a construction based on geohash aimed at trajectory fingerprinting. We demonstrate that by relying on the properties of a space filling curve geodabs can be used to build sharded inverted indexes. We show how normalization affects precision and recall, two key measures in information retrieval. We then demonstrate that the probabilistic nature of fingerprinting has a marginal effect on the quality of the results. Finally, we evaluate our method in terms of performances and show that, in contrast with existing methods, it is not affected by the density of the trajectory dataset and that it can be efficiently distributed.

Vaibhav Kulkarni, Arielle Moro, Bertil Chapuis et al.

Sharing location traces with context-aware service providers has privacy implications. Location-privacy preserving mechanisms, such as obfuscation, anonymization and cryptographic primitives, have been shown to have impractical utility/privacy tradeoff. Another solution for enhancing user privacy is to minimize data sharing by executing the tasks conventionally carried out at the service providers' end on the users' smartphones. Although the data volume shared with the untrusted entities is significantly reduced, executing computationally demanding server-side tasks on resource-constrained smartphones is often impracticable. To this end, we propose a novel perspective on lowering the computational complexity by treating spatiotemporal trajectories as space-time signals. Lowering the data dimensionality facilitates offloading the computational tasks onto the digital-signal processors and the usage of the non-blocking signal-processing pipelines. While focusing on the task of user mobility modeling, we achieve the following results in comparison to the state of the art techniques: (i) mobility models with precision and recall greater than 80%, (ii) reduction in computational complexity by a factor of 2.5, and (iii) reduction in power consumption by a factor of 0.5. Furthermore, our technique does not rely on users' behavioral parameters that usually result in privacy-leakage and conclusive bias in the existing techniques. Using three real-world mobility datasets, we demonstrate that our technique addresses these weaknesses while formulating accurate user mobility models.