Year: 2023
Type: dissertation
Abstract: Synthetic trajectory generation is crucial in order to study and analyse many fields, such as urban mobility, autonomous driving, mobile communication protocols, computational epidemiology and urban planning. Unfortunately, generating synthetic trajectory data is not an easy task, as it is a highly correlated, sequential type of data, with variable length. Currently, some work has been done on the field of synthetic trajectory generation, although none of them take into account the privacy of the user that has provided the trajectories. Furthermore, generating synthetic data without any security concern can put at risk users? privacy, since deep learning models that generate synthetic trajectories are susceptible to membership inference attacks and model inversion attacks. In this project we present an update of an existing deep learning model called MoveSim, which generates synthetic trajectory data. This model has been modified in order to achieve ?-Differential Privacy, and in this way guarantee privacy to the users that have provided their data without losing performance on the model.
Author: Rubio Jornet, Víctor
Institution: Universitat Politècnica de Catalunya
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7000684642
