The widespread use of automated decision processes in many areas of our society raises serious ethical issues with respect to the fairness of the process and the possible resulting discrimination. To solve this issue, we propose a novel adversarial training approach called GANSan for learning a sanitizer whose objective is to prevent the possibility of any discrimination (i.e., direct and indirect) based on a sensitive attribute by removing the attribute itself as well as the existing correlations with the remaining attributes. Our method GANSan is partially inspired by the powerful framework of generative adversarial networks (in particular Cycle-GANs), which offers a flexible way to learn a distribution empirically or to translate between two different distributions. In contrast to prior work, one of the strengths of our approach is that the sanitization is performed in the same space as the original data by only modifying the other attributes as little as possible, thus preserving the interpretability of the sanitized data. Consequently, once the sanitizer is trained, it can be applied to new data locally by an individual on their profile before releasing it. Finally, experiments on real datasets demonstrate the effectiveness of the approach as well as the achievable trade-off between fairness and utility.

The pervasive integration of GPS-enabled devices and population growth, alongside diverse transportation options, has led to an exponential increase in the collection of trajectory data, driving advancement in spatiotemporal data mining and urban planning. However, sharing trajectory data that includes sensitive user information can significantly threaten personal privacy, as location-based services like mobile apps and vehicle tracking offer valuable insights into travel routes while simultaneously raising serious concerns about the exposure of personal geolocation information. To solve this dilemna, we propose a novel generative model based on a spatiotemporal diffusion probabilistic model: Trajectory Denoising Diffusion Generative Adversarial Network (TrajDD-GAN). TrajDD-GAN reconstructs and synthesizes geographic trajectories through a forward diffusion process that adds Gaussian noise to real trajectory data, followed by a multimodal conditional GAN for iterative denoising. This process significantly reduces the number of required diffusion steps. Our experiments demonstrate the potential of integrating probabilistic diffusion techniques with generative adversarial networks to achieve better sample quality, diversity and efficiency for real-world applications. Extensive experimental results show that TrajDD-GAN generates synthetic trajectories that preserve the statistical properties of the original mobility data. One of TrajDD-GAN’s main advantages compared to other diffusion models is its cost-effective inference process: fewer steps mean faster generation. This allows for the efficient generation of large amounts of synthetic data. By optimizing the denoising steps and integrating advanced generative techniques, these models significantly reduce computational overhead while maintaining high-quality output, making them suitable for large-scale applications in trajectory generation and analysis.

This letter presents an approach to over-approximate the reachable set of states of a system whose uncertainties are arbitrarily time-varying. Most approaches generally assume piecewise continuity or sometimes Riemann-integrability of the uncertainties. In this letter we go one step further, only assuming Lebesgue measurability, which is the weakest meaningful hypothesis. We develop our new technique, based on a decomposition of components as a difference of positive functions, for separable systems, a generalization of control-affine systems. We compare the over-approximation produced by our method with the ones obtained using the tools Flow* and CORA on simple examples, and show that correct outer-approximations of the reachable sets are computable with a high degree of precision even for these general forms of uncertainties.

The widespread use of automated decision processes in many areas of our society raises serious ethical issues concerning the fairness of the process and the possible resulting discriminations. In this work, we propose a novel approach called GANsan whose objective is to prevent the possibility of any discrimination i.e., direct and indirect) based on a sensitive attribute by removing the attribute itself as well as the existing correlations with the remaining attributes. Our sanitization algorithm GANsan is partially inspired by the powerful framework of generative adversarial networks (in particular the Cycle-GANs), which offers a flexible way to learn a distribution empirically or to translate between two different distributions. In contrast to prior work, one of the strengths of our approach is that the sanitization is performed in the same space as the original data by only modifying the other attributes as little as possible and thus preserving the interpretability of the sanitized data. As a consequence, once the sanitizer is trained, it can be applied to new data, such as for instance, locally by an individual on his profile before releasing it. Finally, experiments on a real dataset demonstrate the effectiveness of the proposed approach as well as the achievable trade-off between fairness and utility.

Analyse d'atteignabilité avec incertitudes variables dans le temps et intégrables au sens de Lebesgue: L'analyse d'atteignabilité, technique calculant l'ensemble des états atteignables d'un système, permet de prouver des propriétés de sûreté. Ce calcul exact étant impossible dans le cas général, nous calculons des sur-approximations, i.e. ensembles garantis de contenir l'ensemble des états atteignables. Ces calculs sont complexes à cause d'incertitudes variant dans le temps qui permettent de modéliser l'environnement dans lequel évolue le système étudié. Dans ce travail, nous nous intéressons à la gestion d'incertitudes bornées intégrables au sens de Lebesgue, qui le cas le plus général étudié dans l'état de l'art. Certaines méthodes permettent déjà de les prendre en compte mais elles produisent généralement de larges sur-approximations lorsque les ensembles de valeurs des incertitudes sont grands.Nous prouvons que, sous certaines hypothèses, des fonctions ensemblistes contractées par une version ensembliste de l'opérateur de Picard associé à un problème de Cauchy sont des sur-approximations des ensembles d'états atteignables en fonction du temps. Cela prouve en particulier que des intervalles contractés par un tel opérateur sont des sur-approximations valides dans le cas général d'incertitudes bornées intégrables au sens de Lebesgue.Nous proposons ensuite un nouvel algorithme basé sur une décomposition en fonctions positives des équations différentielles dans le cas des systèmes que nous nommons "séparables par rapport aux incertitudes variables dans le temps", une généralisation des systèmes affines par rapport aux contrôles. Un prototype illustre la précision de cet algorithme par rapport à ceux implémentés dans les outils de l'état de l'art.Enfin, nous présentons les implications de l'adaptation d'outils de simulations de systèmes hybrides pour effectuer des analyses d'atteignabilité. Nous nous concentrons sur le langage Zélus qui permet de modéliser des systèmes déterministes en exécutant les dynamiques discrètes aux instants de changement de signe de signaux continus. Nous proposons une paramétrisation des modèles intermédiaires générés par son compilateur et nous illustrons la faisabilité de cette méthode avec un prototype exécuté sur un modèle intermédiaire généré manuellement.