Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 1
Encontrada(s) 1 página(s)

Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

On the adaptation of trajectory generative models to enforce differential privacy through gradient perturbation

  • Rubio Jornet, Víctor
  • Parra Arnau, Javier
  • Forné Muñoz, Jorge
Synthetic trajectories have become a crucial tool for enabling data sharing in domains such as telecommunications and urban planning, yet most generators provide neither formal privacy guarantees nor empirical validation of the protection they claim. This paper investigates whether state-of-the-art deep learning trajectory generators can be adapted to satisfy differential privacy (DP) while retaining usable utility, and makes three contributions. First, we analyze 16 state-of-the-art generators and determine which can be trained with Differentially Private Stochastic Gradient Descent (DP-SGD), showing that common architectural choices — such as index-based coordinate look-ups at inference — silently void the guarantee. Second, we present Sec-LSTM-TrajGAN, a DP adaptation of LSTM-TrajGAN, a well-known trajectory generator chosen after the analysis of the existing models. Third, we introduce RouteMIA, to our knowledge the first closed-box membership inference attack tailored to synthetic trajectories, which adapts the distanceto-closest-record attack to sequential spatio-temporal data and requires access only to the released dataset. Evaluating on GeoLife (17, 000 trajectories) and Porto (100, 000 trajectories) datasets for several scenarios of increasing privacy. RouteMIA attains an AUC of 0.59 and 0.55 against the non-private models, which drops to ˜ 0.50 — chance level — for every private configuration. Notably, gradient clipping alone, with no noise added, already accounts for this entire empirical gain, at a utility cost of a distance JSD rising from 0.12 to 0.20 on Porto and essentially none on GeoLife, whereas strict budgets (e = 3) degrade GeoLife markedly (distance JSD up to 0.35). Per-example clipping, not the noise, therefore delivers the measurable protection against this attack — a cheap empirical complement to, but not a replacement for, formal guarantees., This work was supported in part by Spanish Government, funded by Ministerio de Ciencia e Innovación (MCIN)/Agencia Estatal de Investigación (AEI)/10.13039/501100011033 through the research project ‘‘DIstributed Smart COmmunications with Verifiable EneRgy-optimal Yields (DISCOVERY)’’ under Grant PID2023-148716OB-C32, through the research project ‘‘Privacy-Preserving Techniques for Mobility Data (PRIMOD)’’ under Grant CNS2025-166870, through the research project ‘‘Anonymization Technology for Artificial Intelligence (AI)-Based Analytics of Mobility Data (MOBILYTICS)’’ under Grant TED2021-129782B-I00, and through the research project ‘‘Enhancing Communication Protocols with Machine Learning while Protecting Sensitive Data (COMPROMISE)’’ under Grant PID2020-113795RB-C31; in part by the project MOBILYTICS funded by European Union—NextGenerationEU/Recovery, Transformation and Resilience Plan (PRTR); and in part by the Generalitat de Catalunya through AGAUR under Grant 2021 SGR 01413. The work of Víctor Rubio-Jornet was supported by the Joan Oró Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR)-Formació d’Investigadors (FI) Predoctoral Program, funded by the Secretariat of Universities and Research of the Department of Research and Universities of the Generalitat of Catalonia and European Social Fund Plus under Grant 2024 FI-1 00762. The work of Javier Parra-Arnau was supported by Ramón y Cajal Grant through Spanish Ministry of Science and Innovation and European Union—‘‘NextGenerationEU’’/PRTR under Grant RYC2021-034256-I., Peer Reviewed, Postprint (published version)