International audience

Sensor networks, especially when deployed in a field with little supervision, are vulnerable to a broad range of attacks. In this paper, we study a scenario where multiple competitive adversaries inject false content in the sensed data with the intent of impairing network control. We use game theory to analyze the different behavior of adversaries acting independently or in a coordinated fashion. This analysis ultimately results in the evaluation of efficiency metrics for the utility of uncoordinated attackers, based on the Age of Incorrect Information (AoII), which is compared to the coordinated case. Our numerical results show that generally the lack of coordination is detrimental for the two attackers. With the exception of few edge cases, competition leads the attackers to be more concerned with prevailing over each other than actually compromising the system.

The Metaverse is an immersive online world, accessed through headsets, seamlessly integrating virtual and augmented reality. Users navigate this digital realm through avatars, participating in real-time activities such as work, meetings, concerts. The real-time nature of the Metaverse prompts an analysis using age of information, a metric that tracks information freshness. In this environment, where users actively seek continuous stimuli, sustaining high attention is vital. We propose a game-theoretic analysis of user-server interactions for enduring cooperation, where we incorporate a discount factor to quantitatively compare present and future actions. We derive closed-form solutions for the infinite horizon game and obtain lower bounds for the discount factor chosen by the entities and upper bounds for the communication cost sustainable in order to achieve long-lasting cooperation. This enriches our understanding of temporal dynamics in ensuring information freshness, providing insight into the dynamic interplay between users and the Metaverse environment.

In this paper, we evaluate the impact of local differential privacy (LDP) on the utility of human mobility data obtained from mobile location services. Specifically, we focus our study on visit data, which consist of user-level information on visited locations. This includes the duration of each visit and its category, such as restaurant or department store. The purpose of LDP is to protect sensitive information in visit records by introducing properly calibrated noise, while still allowing the extraction of useful statistics. To evaluate our approach, we study how different levels of privacy budget ϵ impact the utility of the data. The utility is determined by the estimation accuracy for different statistics of interest, such as the number of visits and the average visit duration for each category. We conduct our evaluation on a visits dataset including records from over 20 million mobile devices. Our findings indicate that the number of visits to popular categories can be accurately estimated even at strong privacy levels (ϵ = 1). The estimation of the average visit duration is generally less precise, but it remains feasible under less stringent privacy levels (ϵ ≥ 2 or ϵ ≥ 4, depending on the application at hand).

Recently, there has been a significant surge of lifelogging experiments, where the activity of few participants\nis monitored for a number of days through fitness trackers. Data from such experiments can be aggregated\nin datasets and released to the research community. To protect the privacy of the participants, fitness datasets\nare typically anonymized by removing personal identifiers such as names, e-mail addresses, etc. However,\nalthough seemingly correct, such straightforward approaches are not sufficient. In this paper we demonstrate\nhow an adversary can still de-anonymize individuals in lifelogging datasets. We show that users’ privacy can\nbe compromised by two approaches: (i) through the inference of physical parameters such as gender, height,\nand weight; and/or (ii) via the daily routine of participants. Both methods rely solely on fitness data such as\nsteps, burned calories, and covered distance to obtain insights on the users in the dataset. We train several\ninference models, and leverage them to de-anonymize users in public lifelogging datasets. Between our two\napproaches we achieve 93.5% re-identification rate of participants. Furthermore, we reach 100% success rate\nfor people with highly distinct physical attributes (e.g., very tall, overweight, etc.).

Growing privacy concerns regarding personal data disclosure are contrasting with the constant need of such information for data-driven applications. To address this issue, the combination of federated learning and differential privacy is now well-established in the domain of machine learning. These techniques allow to train deep neural networks without collecting the data and while preventing information leakage. However, there are many scenarios where simpler and more robust machine learning models are preferable. In this paper, we present a federated and differentially-private version of the Naive Bayes algorithm for classification. Our\nresults show that, without data collection, the same performance of a centralized solution can be achieved on any dataset with only a slight increase in the privacy budget. Furthermore, if certain conditions are met, our federated solution can outperform a centralized approach.

We explore the threat posed by disclosure of personal fitness information collected by wearable devices. In\nparticular, we study a scenario where an attacker has a list of aggregated records produced by a group of users,\nwhich are stored as time series of steps and calories. We introduce a machine learning-based approach to\nidentify one target person in the aggregated data while being in possession of other records from that person.\nWe estimate how accurately an attacker can find the target’s data when aggregated with other users by testing\nour approach on two public datasets. Our results show that personal fitness data possess identifying capabilities\nthat should be accounted when they are shared or disclosed.

We propose a data-driven secure wireless communication scheme, in which the goal is to transmit a signal to a legitimate receiver with minimal distortion, while keeping some information about the signal private from an eavesdropping adversary. When the data distribution is known, the optimal trade-off between the reconstruction quality at the legitimate receiver and the leakage to the adversary can be characterised in the information theoretic asymptotic limit. In this paper, we assume that we do not know the data distribution, but instead have access to a dataset, and we are interested in the finite blocklength regime rather than the asymptotic limits. We propose a data-driven adversarially trained deep joint source-channel coding architecture, and demonstrate through experiments with CIFAR-10 dataset that it is possible to transmit to the legitimate receiver with minimal end-to-end distortion while concealing information on the image class from the adversary.

LoRa is a low-power wide-area network solution that is recently gaining popularity in the context of the Internet of Things due to its ability to handle massive number of devices. One of the main challenges faced by LoRa implementations is the allocation of Spreading Factors to the devices. While the assignment of these parameters is virtually simple to execute, scalability and complexity issues hint at its implementation through a game theoretic approach. This would offer the advantage of being readily implementable in vast networks of devices with limited hardware capabilities. Hence, we formulate the SF allocation problem as a Bayesian game, of which we compute the Bayesian Nash equilibria. We also implement the procedure in the ns- 3 network simulator and evaluate the resulting performance, showing that our approach is scalable and robust, and also offers room for improvement with respect to existing approaches.