Search

International audience; We consider elastic resource provisioning in the cloud, focusing on in-memory key-value stores used as caches. Our goal is to dynamically scale resources to the traffic pattern minimizing the overall cost, which includes not only the storage cost, but also the cost due to misses. In fact, a small variation of the cache miss ratio may have a significant impact on user perceived performance in modern web services, which in turn has an impact on the overall revenues for the content provider using such services. We propose and study a dynamic algorithm for TTL caches, which is able to obtain close-to-minimal costs. Since high-throughput caches require low complexity operations, we discuss a practical implementation of such a scheme requiring constant overhead per request independently from the cache size. We evaluate our solution with real-world traces collected from Akamai, and show that the TTL approach is able to track the optimal cache configuration and achieve significant cost savings specially in highly dynamic settings that are likely to require elastic cloud services.

International audience; Résumé de [Canale et al. (2018)], publié à ISWC 2018. Une tâche cruciale en extraction de connaissances à partir de textes se décompose souvent en deux tâches complémentaires : la reconnaissance d'entité nommée (NER) et la désambiguïsation d'entité nommée (NED). L'objectif consiste à attribuer à des parties du texte (mention) respectivement un type appartenant à une taxonomie prédéfinie et un identifiant unique, souvent représenté sous la forme d'URI, qui fait référence de manière univoque à une entité définie dans une base de connaissances donnée. La combinaison de ces deux tâches est souvent abrégée avec l'acronyme NERD. De nombreuses approches, souvent exposées sous forme d'API Web, ont été proposées pour ré-soudre ces tâches au cours des dernières années. En termes de NER, chaque service fournit générale-ment sa propre taxonomie de types qui peuvent être reconnus. Même si tous comprendront trois types principaux (PERSON, ORGANIZATION, LOCATION), ils diffèrent largement pour les types plus fins, ce qui complique leur comparaison et leur combinaison. En termes de NED, chaque extracteur peut potentiellement lever l'ambiguïté d'entités par rapport à des bases de connaissances spécifiques (KB), mais en pratique, ils s'appuient principalement sur des bases de connaissances généralistes, comme DBpedia ou Wikidata. Pour cette raison, la comparaison et la fusion des résultats de ces ex-tracteurs nécessitent certaines tâches de post-traitement qui dépendent généralement d'alignements entre ces bases de connaissances. Dans ce travail, nous décrivons Ensemble NERD, un framework qui regroupe de nombreuses réponses d'extracteurs, les normalise et les combine afin de produire des annotations sémantiques. Cette méthode repose sur deux réseaux d'apprentissage profond, ENNTR (Ensemble Neural Network for Type Recognition) et ENND (Ensemble Neural Network for Disambiguation), qui fournissent des modèles pour effectuer d'une part un alignement entre les types et d'autre part entre les entités nommées identifiées dans une base de connaissances.

International audience; This paper presents a novel dataset for image and face manipulation detection and localization called DEFACTO. The dataset was automatically generated using Microsoft common object in context database (MSCOCO) to produce semantically meaningful forgeries. Four categories of forgeries have been generated. Splicing forgeries which consist of inserting an external element into an image, copy-move forgeries where an element within an image is duplicated, object removal forgeries where objects are removed from images and lastly morphing where two images are warped and blended together. Over 200000 images have been generated and each image is accompanied by several annotations allowing precise localization of the forgery and information about the tampering process.

In modern cities, a rapid increase of motorcycles and other types of Powered Two-Wheelers (PTWs) is observed as an answer to long commuting in traffic jams and complex urban navigation. Such increasing penetration rate of PTWs creates mixed traffic flow conditions with unique characteristics that are not well understood at present. Our objective is to develop an analytical traffic flow model that reflects the mutual impacts of PTWs and Cars. Unlike cars, PTWs filter between cars, have unique dynamics, and do not respect lane discipline, therefore requiring a different modeling approach than traditional “Passenger Car Equivalent” or “Follow the Leader”. Instead, this work follows an approach that models the flow of PTWs similarly to a fluid in a porous medium, where PTWs “percolate” between cars depending on the gap between them. Our contributions are as follows: (I) a characterization of the distribution of the spacing between vehicles by the densities of PTWs and cars; (II) a definition of the equilibrium speed of each class as a function of the densities of PTWs and cars; (III) a mathematical analysis of the model’s properties (IV) an impact analysis of the gradual penetration of PTWs on cars and on heterogeneous traffic flow characteristics. The proposed model could contribute as an enabler for ‘PTW-aware’ future Cooperative Intelligent Transport Systems technologies and traffic regulations.

Malware is still a widespread problem, and it is used by malicious actors to routinely compromise the security of computer systems. Consumers typically rely on a single AV product to detect and block possible malware infections, while corporations often install multiple security products, activate several layers of defenses, and establish security policies among employees. However, if a better security posture should lower the risk of malware infections, then the actual extent to which this happens is still under debate by risk analysis experts. Moreover, the difference in risks encountered by consumers and enterprises has never been empirically studied by using real-world data. In fact, the mere use of third-party software, network services, and the interconnected nature of our society necessarily exposes both classes of users to undiversifiable risks: Independently from how careful users are and how well they manage their cyber hygiene, a portion of that risk would simply exist because of the fact of using a computer, sharing the same networks, and running the same software. In this work, we shed light on both systemic (i.e., diversifiable and dependent on the security posture) and systematic (i.e., undiversifiable and independent of the cyber hygiene) risk classes. Leveraging the telemetry data of a popular security company, we compare, in the first part of our study, the effects that different security measures have on malware encounter risks in consumer and enterprise environments. In the second part, we conduct exploratory research on systematic risk, investigate the quality of nine different indicators we were able to extract from our telemetry, and provide, for the first time, quantitative indicators of their predictive power. Our results show that even if consumers have a slightly lower encounter rate than enterprises (9.8% vs. 12.0%), the latter do considerably better when selecting machines with an increasingly higher uptime (89% vs. 53%). The two segments also diverge when we separately consider the presence of Adware and Potentially Unwanted Applications (PUA) and the generic samples detected through behavioral signatures: While consumers have an encounter rate for Adware and PUA that is 6 times higher than enterprise machines, those on average match behavioral signatures 2 times more frequently than the counterpart. We find, instead, similar trends when analyzing the age of encountered signatures, and the prevalence of different classes of traditional malware (such as Ransomware and Cryptominers). Finally, our findings show that the amount of time a host is active, the volume of files generated on the machine, the number and reputation of vendors of the installed applications, the host geographical location, and its recurrent infected state carry useful information as indicators of systematic risk of malware encounters. Activity days and hours have a higher influence in the risk of consumers, increasing the odds of encountering malware of 4.51 and 2.65 times. In addition, we measure that the volume of files generated on the host represents a reliable indicator, especially when considering Adware. We further report that the likelihood of encountering Worms and Adware is much higher (on average 8 times in consumers and enterprises) for those machines that already reported this kind of signature in the past.

The recent introduction of vector coded caching has revealed that multi-rank transmissions in the presence of receiver-side cache content can dramatically ameliorate the file-size bottleneck of coded caching and substantially boost performance in error-free wire-like channels. In this work, we employ large-matrix analysis to explore the effect of vector coded caching in realistic wireless multi-antenna downlink systems. For a given downlink MISO system already optimized to exploit both multiplexing and beamforming gains, and for a fixed set of antenna and SNR resources, our analysis answers a simple question: What is the multiplicative throughput boost obtained from introducing reasonably sized receiver-side caches that can pre-store information content? The derived closed-form expressions capture various linear precoders, and a variety of practical considerations such as power dissemination across signals, realistic SNR values, as well as feedback costs. The schemes are very simple (we simply collapse precoding vectors into a single vector), and the recorded gains are notable. For example, for 32 transmit antennas, a received SNR of 20 dB, a coherence bandwidth of 300 kHz, a coherence period of 40 ms, and under realistic file-size and cache-size constraints, vector coded caching is here shown to offer a multiplicative throughput boost of about 310% with ZF/RZF precoding and a 430% boost in the performance of already optimized MF-based (cacheless) systems. Interestingly, vector coded caching also accelerates channel hardening to the benefit of feedback acquisition, often surpassing 540% gains over traditional hardening-constrained cacheless downlink systems. Comunidad de Madrid - Atracción de talento TRUE pub

This work identifies the fundamental limits of Multi-Access Coded Caching (MACC) where each user is connected to multiple caches in a manner that follows a generalized combinatorial topology. This topology stands out as it allows for unprecedented coding gains, even with very modest cache resources. First, we extend the setting and the scheme presented by Muralidhar et al. to a much more general topology that supports both a much denser range of users and the coexistence of users connected to different numbers of caches, all while maintaining the astounding coding gains, here proven to be exactly optimal, associated with the combinatorial topology. This is achieved, for this generalized topology, with a novel information-theoretic converse that we present here, which establishes, together with the scheme, the exact optimal performance under the assumption of uncoded placement. We subsequently consider different connectivity ensembles, including the very general scenario of the entire ensemble of all possible network connectivities/topologies, where any user can be connected to any subset of caches arbitrarily. For these settings, we develop novel converse bounds on the optimal performance averaged over the ensemble's different connectivities, considering the additional realistic scenario that the connectivity at delivery time is entirely unknown during cache placement. This novel analysis of topological ensembles leaves open the possibility that currently-unknown topologies may yield even higher gains, a hypothesis that is part of the bigger question of which network topology yields the most caching gains., 48 pages, 2 figures. Submitted to IEEE Transactions on Information Theory

Coordinated vehicle control strategies aim at optimizing driving dynamics to increase traffic flow without impacting safety. These control strategies are based on the knowledge of the vehicles' state information like position and velocity obtained through Vehicle-to-everything (V2X) communications. Literature on control strategies yet assumes perfect positions, whereas position errors are in fact present and non negligible (e.g. GPS). As a consequence, these localization errors impact the control strategies by introducing uncertainty, which must be accounted for to minimize the probability of accidents. This paper qualifies and quantifies such uncertainty and proposes strategies to reduce it in a collision avoidance scenario. We notably relate these strategies to their impacts on traffic flow. More specifically, we model coordinated automated vehicles as a Model Predictive Control (MPC), integrate localization errors and evaluate its impact of the output to avoid accident. We then propose possibilities to mitigate accident-prone controls and quantify them on traffic flow. Our study illustrates that localization errors impact traffic flow by forcing future automated vehicles to increase gaps or reduce speed.

Human factors include human errors and human weaknesses such as long reaction time, limited visibility, limited jerk sustainability, etc. Such human factors account for a large portion of vehicular accidents involving manually driven vehicles (MDVs). Future Intelligent Vehicles (IVs) are expected to have higher perception ranges, faster reactions and coordinated mobility capability. At early market penetration, IVs will share road with MDVs and will adjust its driving dynamics according to the dynamics of other neighboring vehicles. In such a scenario, we investigate how IVs can compensate for human factors impacting MDVs and accordingly improve traffic safety in mixed traffic conditions. We model a centralized Model Predictive Control (MPC) based controller and integrate constraints derived from human factors. We show that IVs may not only reduce collisions, but also increase traffic density without negatively impacting traffic safety. We finally assess the impact of a gradual penetration of IVs in a mixed vehicle scenario.

International audience

International audience

Searchable Encryption (SE) allows a user to upload data to the cloud and to search it in a remote fashion while preserving the privacy of both the data and the queries. Recent research results describe attacks on SE schemes using the access pattern, denoting the ids of documents matching search queries, which most SE schemes reveal during query processing. However SE schemes usually leak more than just the access pattern, and this extra leakage can lead to attacks (much) more harmful than the ones using basic access pattern leakage only. We remark that in the special case of Multi-User Searchable Encryption (MUSE), where many users upload and search data in a cloud-based infrastructure, a large number of existing solutions have a common leakage in addition to the well-studied access pattern leakage. We show that this seemingly small extra leakage allows a very simple yet powerful attack, and that the privacy degree of the affected schemes have been overestimated. We also show that this new vulnerability affects existing software. Finally we formalize the newly identified leakage profile and show how it relates to previously defined ones.

International audience; Body area networks (BANs), referring to embedded wireless systems in, on and around bodies, are expected to take an important role for health, leisure, sports and all facets of our daily life. In many situations, several BANs coexist in a small environment, resulting in very strong inter-BAN interference which may seriously disturb communications in these BANs. The goal of this paper is to decrease inter-BAN interference by cooperative scheduling, hence increasing packet reception rate (PRR) of intra-BAN communications. The scheduling issue is divided into two sub-issues: single- BAN scheduling as an optimization issue and multi-BAN concurrent scheduling as a game. For the first sub-issue, horse racing scheduling is proposed, which achieves near-optimal PRR for the scheduling BAN with low complexity. For the second sub-issue, we prove the existence of a set of mixed strategy Nash equilibria (NE) then propose a distributed cooperative scheduling scheme, which efficiently achieves higher PRR than the always existing MSNE without degrading fairness.

International audience; Given the growing adoption of AI, cloud data lakes are facing the need to support cost-effective "just-in-case" data archival over long time periods to meet regulatory compliance requirements. Unfortunately, current media technologies suffer from fundamental issues that will soon, if not already, make cost-effective data archival infeasible. In this paper, we present a vision for redesigning the archival tier of cloud data lakes based on a novel, obsolescence-free storage medium-synthetic DNA. In doing so, we make two contributions: (i) we highlight the challenges in using DNA for data archival and list several open research problems, (ii) we outline OligoArchive-DSM (OA-DSM)-an end-to-end DNA storage pipeline that we are developing to demonstrate the feasibility of our vision.

We present Malafide, a universal adversarial attack against automatic speaker verification (ASV) spoofing countermeasures (CMs). By introducing convolutional noise using an optimised linear time-invariant filter, Malafide attacks can be used to compromise CM reliability while preserving other speech attributes such as quality and the speaker's voice. In contrast to other adversarial attacks proposed recently, Malafide filters are optimised independently of the input utterance and duration, are tuned instead to the underlying spoofing attack, and require the optimisation of only a small number of filter coefficients. Even so, they degrade CM performance estimates by an order of magnitude, even in black-box settings, and can also be configured to overcome integrated CM and ASV subsystems. Integrated solutions that use self-supervised learning CMs, however, are more robust, under both black-box and white-box settings., Accepted at INTERSPEECH 2023

International audience; A mobile banking app just started up, and the notification "App updated, click here to restart" appears. The graphic theme is the same as the bank. Can we trust it? What if we cannot even trust that tapping an app actually loads the original one? More generally, what if Android notifies an attacker when her victim has just launched the target app of her phishing campaign so that she could cast the hook at the perfect moment? In this paper, we abuse inotify APIs, a mechanism for monitoring file system events, to mount a state inferencebased phishing attack from a malicious app installed on the victim's smartphone. We also verified the novelty of our work analyzing 10,000 recent Android malware, and although we found some cases where malware uses inotify for their petty purposes, our attack seems to be publicly unknown. However, since Android constantly evolves year after year, we studied its feasibility over different Android versions and attacker's capabilities. By analyzing 4,863 of the most popular apps, the most disconcerting finding is that if the attacker knows the installation path of the target app, all Android apps are vulnerable, regardless of the system version. Getting the installation path of an app is a capability that is only protected by a normal permission, and to make matters worse, there are workarounds to get it even without such permission. Even if this capability is denied, we propose different attack models under which this attack is still possible; however, at the end of our work, we provide the remediation to eradicate once and for all these attacks. Through this work, we reported three vulnerabilities to Google. Two were acknowledged as bugs of moderate severity, while the last one was already known but not public.

This paper is co-authored by an informal group of experts from a broad range of backgrounds all of whom are active in standards groups, consortia and/or alliances in the Internet of Things (IoT) space. The ambition is to create mindshare on approaches to semantic interoperability and to actively encourage consensus building on what the co-authors regard as a key technical issue

International audience; no abstract

The first step required to perform any analysis of a physical memory image is the reconstruction of the virtual address spaces, which allows translating virtual addresses to their corresponding physical offsets. However, this phase is often overlooked, and the challenges related to it are rarely discussed in the literature. Practical tools solve the problem by using a set of custom heuristics tailored on a very small number of well-known operating systems (OSs) running on few architectures. In this article, we look for the first time at all the different ways the virtual to physical translation can be operated in 10 different CPU architectures. In each case, we study the inviolable constraints imposed by the memory management unit that can be used to build signatures to recover the required data structures from memory without any knowledge about the running OS. We build a proof-of-concept tool to experiment with the extraction of virtual address spaces showing the challenges of performing an OS-agnostic virtual to physical address translation in real-world scenarios. We conduct experiments on a large set of 26 different OSs and a use case on a real hardware device. Finally, we show a possible usage of our technique to retrieve information about user space processes running on an unknown OS without any knowledge of its internals.

Over the last decade, the use of Automatic Speaker Verification (ASV) systems has become increasingly widespread in response to the growing need for secure and efficient identity verification methods. The voice data encompasses a wealth of personal information, which includes but is not limited to gender, age, health condition, stress levels, and geographical and socio-cultural origins. These attributes, known as soft biometrics, are private and the user may wish to keep them confidential. However, with the advancement of machine learning algorithms, soft biometrics can be inferred automatically, creating the potential for unauthorized use. As such, it is crucial to ensure the protection of these personal data that are inherent within the voice while retaining the utility of identity recognition. In this paper, we present an adversarial Auto-Encoder--based approach to hide gender-related information in speaker embeddings, while preserving their effectiveness for speaker verification. We use an adversarial procedure against a gender classifier and incorporate a layer based on the Laplace mechanism into the Auto-Encoder architecture. This layer adds Laplace noise for more robust gender concealment and ensures differential privacy guarantees during inference for the output speaker embeddings. Experiments conducted on the VoxCeleb dataset demonstrate that speaker verification tasks can be effectively carried out while concealing speaker gender and ensuring differential privacy guarantees; moreover, the intensity of the Laplace noise can be tuned to select the desired trade-off between privacy and utility., 6 Pages, 3 figures

International audience; This paper studies a novel multi-access coded caching (MACC) model where the topology between users and cache nodes is a generalization of those already studied in previous work, such as combinatorial and cross-resolvable design topologies. Our goal is to minimize the worst-case transmission load in the delivery phase from the server over all possible user requests. By formulating the access topology as two classical combinatorial structures, t-design and t-group divisible design, we propose two classes of coded caching schemes for a flexible number of users, where the number of users can scale linearly, polynomially or exponentially with the number of cache nodes. In addition, our schemes can unify most schemes for the shared link network and unify many schemes for the multi-access network except for the cyclic wrap-around topology.

Spoofing countermeasure (CM) and automatic speaker verification (ASV) sub-systems can be used in tandem with a backend classifier as a solution to the spoofing aware speaker verification (SASV) task. The two sub-systems are typically trained independently to solve different tasks. While our previous work demonstrated the potential of joint optimisation, it also showed a tendency to over-fit to speakers and a lack of sub-system complementarity. Using only a modest quantity of auxiliary data collected from new speakers, we show that joint optimisation degrades the performance of separate CM and ASV sub-systems, but that it nonetheless improves complementarity, thereby delivering superior SASV performance. Using standard SASV evaluation data and protocols, joint optimisation reduces the equal error rate by 27\% relative to performance obtained using fixed, independently-optimised sub-systems under like-for-like training conditions., Comment: Accepted to ICASSP 2023. Code will be available soon

International audience; We propose a geographic content placement policy for wireless caching networks. This policy, named Joint Caching Policy (JCP), jointly determines the caching strategy across the set of base stations (BSs) to improve the hit probability of an arbitrarily located user in the network versus caching policies where placement is independent and identically distributed across the BSs. To that end, JCP divides the BSs into groups, and executes a joint caching policy for each group, while content placement is independent across the groups. Existing joint caching policies require knowledge of user location to optimize content placement. On the other hand, JCP does not require any information on user location, and it provides a content placement policy that outperforms the one given by Independent Caching Policy (ICP), proposed by Błaszczyszyn and Giovanidis in 2015, under any user location distribution. We prove that the hit probability under JCP is lower bounded by that of ICP. We further propose an extension of JCP, named JCP-OPT, which improves the hit probability over JCP by solving a concave maximization problem, provided that there is side information about the user locations. We validate the performance of JCP and JCP-OPT via numerical evaluations and demonstrate that they can provide up to a 30% gain in hit probability over ICP.

Accepted paper to ISIT 2023 without proofs; International audience; In this paper, we study the computation of the ratedistortion-perception function (RDPF) for discrete memoryless sources subject to a single-letter average distortion constraint and a perception constraint that belongs to the family of fdivergences. For that, we leverage the fact that RDPF, assuming mild regularity conditions on the perception constraint, forms a convex programming problem. We first develop parametric characterizations of the optimal solution and utilize them in an alternating minimization approach for which we prove convergence guarantees. The resulting structure of the iterations of the alternating minimization approach renders the implementation of a generalized Blahut-Arimoto (BA) type of algorithm infeasible. To overcome this difficulty, we propose a relaxed formulation of the structure of the iterations in the alternating minimization approach, which allows for the implementation of an approximate iterative scheme. This approximation is shown, via the derivation of necessary and sufficient conditions, to guarantee convergence to a globally optimal solution. We also provide sufficient conditions on the distortion and the perception constraints which guarantee that our algorithm converges exponentially fast. We corroborate our theoretical results with numerical simulations, and we draw connections with existing results.

International audience; The zero Touch Management (ZSM) concept in 5G and Beyond networks (B5G) aims to automate the management and orchestration of running network slices. This requires heavy usage of advanced deep learning techniques in a closed-loop way to auto-build the suitable decisions, enabling to meet network slices' requirements. In this context, Federated Learning (FL) is playing a vital role in training deep learning models in a collaborative way among thousands of network slice participants while ensuring their privacy and hence network slice isolation. Specifically, running network slices may share only their model parameters with a central entity, e.g., Inter Domain Slice Manager, to aggregate them and build a global model. Thus, the central entity does not directly access the training data. However, FL is vulnerable to poisoning attacks, where an insider participant may upload poisoning updates to the central entity so that it can cause a construction failure of the global model and thus affect its global performance. Therefore, it is crucial to design security means to detect and mitigate such threats. In this paper, we design a novel framework to automatically detect malicious participants in the FL process. In particular, our framework first uses a deep reinforcement algorithm to dynamically select a network slice as a trusted participant, based mainly on its reputation. The selected participant will then be in charge of identifying poisoning model updates by leveraging unsupervised machine learning. We demonstrate the feasibility of our framework on top of a real dataset that we generate using the 5G OpenAirInterface (OAI) platform. Evaluation results show the efficiency of our framework in dealing with poisoning attacks even with the presence of several malicious participants.

We study a variant of a robust description source coding framework via its corresponding characterization, which is a relevant model for goal-oriented semantic information transmission. Considering two individual single-letter separable distortion constraints and input and output data acting as the intrinsic and extrinsic message, respectively, we first derive bounds on the optimal rates of the problem, as well as necessary and sufficient conditions for these bounds to be tight. Subsequently, we prove a general result that provides in parametric form the optimal solution of the characterization of this problem. Capitalizing on these results, we examine the structure of the solution for one case study of general binary alphabets under Hamming distortions and solve in closed form a special case. We also solve another general binary alphabet case where a Hamming and an erasure distortion coexist, as a means to highlight the importance of selecting the type of the distortion constraint in goal-oriented semantic communication. We also develop a semantic-aware Blahut-Arimoto (BA) algorithm, which can be used for the computation of any finite alphabet intrinsic or extrinsic message under individual distortion criteria. Finally, we revisit the problem for multidimensional independent and identically distributed (IID) jointly Gaussian processes with individual mean-square error (MSE) distortion constraints, providing new insights that have previously been overlooked. This work reveals the cardinal role of context-dependent fidelity criteria in goal-oriented semantic communication.

Future communication networks are envisioned to satisfy increasingly granular and dynamic requirements to accommodate the application and user demands. Indeed, novel immersive and mission-critical services necessitate increased computing and network resources, reduced communication latency, and guaranteed reliability. Thus, efficient and adaptive resource management schemes are required to provide and maintain sufficient levels of Quality of Experience (QoE) during the service life-cycle. Service migration is considered a key enabler of dynamic service orchestration. Indeed, moving services on demand is an efficient mechanism for user mobility support, load balancing in case of fluctuations in service demands, and hardware failure mitigation. However, service migration requires planning, as multiple parameters must be optimized to reduce service disruption to a minimum. Recent breakthroughs in computational capabilities allowed the emergence of Machine Learning as a tool for decision making that is expected to enable seamless automation of network resource management by predicting events and learning optimal decision policies. This paper surveys contributions applying Machine Learning (ML) methods to optimize service migration, providing a detailed literature review on recent advances in the field and establishing a classification of current research efforts with an analysis of their strengths and limitations. Finally, the paper provides insights on the main directions for future research.

International audience; This work elevates coded caching networks from their purely information-theoretic framework to a stochastic setting, by exploring the effect of random user activity and by exploiting correlations in the activity patterns of different users. In particular, the work studies the K-user cache-aided broadcast channel with a limited number of cache states (i.e., the content stored at the cache of a certain user), and explores the effect of cache state association strategies in the presence of arbitrary user activity levels; a combination that strikes at the very core of the coded caching problem and its crippling subpacketization bottleneck. We first present a statistical analysis of the average worst-case delay performance of such subpacketization-constrained (state-constrained) coded caching networks, and provide computationally efficient performance bounds as well as scaling laws for any arbitrary probability distribution of the user-activity levels. The achieved performance is a result of a novel user-to-cache state association algorithm that leverages the knowledge of probabilistic user-activity levels. We then follow a data-driven approach that exploits the prior history on user-activity levels and correlations, in order to predict interference patterns, and thus better design the caching algorithm. This optimized strategy is based on the principle that users that overlap more, interfere more, and thus have higher priority to secure complementary cache states. This strategy is proven here to be within a small constant factor from the optimal. Finally, the above analysis is validated numerically using synthetic data following the Pareto principle. To the best of our understanding, this is the first work that seeks to exploit user-activity levels and correlations, in order to map future interference and design optimized coded caching algorithms that better handle this interference.

massive Machine Type Communications (mMTC) network slices in 5G aim to connect a massive number of MTC devices, opening the door for a widened attack surface. Network slices are well isolated, resulting in a low impact on other running slices when attackers control IoT devices belonging to a mMTC network slice (i.e., in-slice attack). However, the impact of the in-slice attacks on the shared infrastructure components with other slices, such as the 5G Core Network (CN), can be harmful, considering the massive number that can be part of mMTC slice. In this paper, we propose a zero-touch security management solution that uses Machine Learning (ML) to detect and mitigate in-slice attacks on 5G CN components, focusing on Distributed Denial of Service (DDoS) attacks. To this aim, we propose: (1) a novel closed-control loop that assists the 5G CN in detecting and mitigating attacks; (2) a ML algorithm that predicts the upper bound of expected MTC devices Attach Requests during a time interval (or an event); (3) a detection algorithm that analyzes an event and uses the ML output to compute a probability that a specific device has participated to an attack; (4) a mitigation algorithm that disconnects and blocks MTC devices suspected to be part of an attack; (5) a Proof of concept implementation on top of a 5G facility.

The real-world implementation of federated learning is complex and requires research and development actions at the crossroad between different domains ranging from data science, to software programming, networking, and security. While today several FL libraries are proposed to data scientists and users, most of these frameworks are not designed to find seamless application in medical use-cases, due to the specific challenges and requirements of working with medical data and hospital infrastructures. Moreover, governance, design principles, and security assumptions of these frameworks are generally not clearly illustrated, thus preventing the adoption in sensitive applications. Motivated by the current technological landscape of FL in healthcare, in this document we present Fed-BioMed: a research and development initiative aiming at translating federated learning (FL) into real-world medical research applications. We describe our design space, targeted users, domain constraints, and how these factors affect our current and future software architecture.

International audience; A new variation of the classical point-to-point joint source-channel coding (JSCC) is studied here, which is relevant for identifying goal-oriented semantic aspects of a transmitted source message over a noisy channel. For this new formulation, coined goal-oriented JSCC, we first introduce optimality criteria followed by necessary and sufficient conditions for global optimality. The focus of our main theoretical results is on investigating the implications of a special subclass of block codes, namely, the single-letter codes, via a theorem in which we provide necessary and sufficient conditions for the probabilistic matching of a goal-oriented source message with a noisy channel. We corroborate our theoretical results with two examples, in which goal-oriented single-letter codes and uncoded transmission perform optimally. Our results further highlight the important role of multiple fidelity constraints in goal-oriented communications.

International audience; This paper investigates when the importance weighting (IW) correction is needed to address covariate shift, a common situation in supervised learning where the input distributions of training and test data differ. Classic results show that the IW correction is needed when the model is parametric and misspecified. In contrast, recent results indicate that the IW correction may not be necessary when the model is nonparametric and well-specified. We examine the missing case in the literature where the model is nonparametric and misspecified, and show that the IW correction is needed for obtaining the best approximation of the true unknown function for the test distribution. We do this by analyzing IW-corrected kernel ridge regression, covering a variety of settings, including parametric and nonparametric models, well-specified and misspecified settings, and arbitrary weighting functions.

International audience; The paper advocates for applying Internet of Things (IoT) for personalized healthcare in smart homes. An IoT architecture is presented which enables such healthcare services. Continuous monitoring of physical parameters and processing of the medical data form the basis of smarter, connected and personalized healthcare. The core functionalities of the IoT architecture are exposed using RESTful web services to the consumers. The main contributions of the paper are: (i) providing a unified approach for discovery, management and interaction with physical devices, (ii) utilizing Machine-to-Machine Measurement (M3) framework to generate high level abstraction from wearable sensors, (iii) combining cross-domain sensor data using M3 and (iv) controlling home automation devices based on health sensor information. Prototype implementations and performance evaluations of the architecture components are discussed.

International audience; This paper presents a user centric and crossdomain IoT application development framework for smart devices. The framework includes automatic discovery of M2M devices, provisioning of sensors and IoT domains, semantic reasoning on sensor data and actuation based on the suggestions. The framework integrates a lightweight version of the Machineto- Machine Measurement framework and has been adapted for smart devices enabling them to reason on sensor data. This allows us to build cross-domain IoT applications for end users. The issues of interoperability, heterogeneity of domains and multimodality of M2M devices are handled by the semantic web technologies. The framework is integrated into oneM2M architecture and its prototype implementation is also presented. The performance of the mobile application is evaluated in terms of CPU load and power consumption. Finally overall contributions and future directions are summarized.

International audience; Semantics is becoming a requirement in Internet of Things (IoT). Recent works underline future challenges for IoT: (1) integrate semantic web technologies, (2) provide interoperability, (3) interpret IoT data, (4) ease the development of IoT applications, and (5) fit to the different requirements of people and applications. In this paper, we design a semantic engine to meet these requirements and integrate it in different components available in IoT architectures. To the best of our knowledge, the main novelty of this paper is the semantic engine flexible enough for different IoT architectures where semantics is integrated in: (1) cloud, (2) devices such as smart phones or tablets, and (3) M2M gateways. As proof of concepts, we discuss deployment of the proposed semantic engine at cloud systems and mobile devices. Moreover, we show that such deployments are coherent with ETSI M2M and oneM2M standardizations.

International audience; An efficient and scalable monitoring system is a critical component for any network to monitor and validate the functioning of the running services and the underlying infrastructure. This is more valid in 5G, as it relies on the network slicing concept, which adds many challenges to the monitoring system. Besides data isolation and multi-tenancy support, network slices require monitoring different types of resources, like RAN, computing, memory, network data rate, which belong to different technological domains, managed by different entities. Moreover, the monitoring system needs to be scalable, as in 5G a high-number of running network slices is envisioned. In this paper, we devise a novel monitoring framework for network slicing ready mobile networks, which features: 1) scalable monitoring system that supports a high number of running network slices in parallel; 2) technological domain agnostic thanks to a novel data collection (or monitoring) communication protocol; 3) support of multi-tenancy in a cloudnative environment. The framework has been implemented in a 5G facility, and its performance has been extensively evaluated.

In this work, we study the spectrum of the normalized Laplacian and its regularized version for random geometric graphs (RGGs) in various scaling regimes. Two scaling regimes are of special interest, the connectivity and the thermodynamic regime. In the connectivity regime, the average vertex degree grows logarithmically in the graph size or faster. In the thermodynamic regime, the average vertex degree is a constant. We introduce a deterministic geometric graph (DGG) with nodes in a grid and provide an upper bound to the probability that the Hilbert–Schmidt norm of the difference between the normalized Laplacian matrices of the RGG and DGG is greater than a certain threshold in both the connectivity and thermodynamic regime. Using this result, we show that the RGG and DGG normalized Laplacian matrices are asymptotically equivalent with high probability (w.h.p.) in the full range of the connectivity regime. The equivalence is even stronger and holds almost surely when the average vertex degree $$a_n$$ a n satisfies the inequality $$a_n > 24 \log (n).$$ a n > 24 log ( n ) . Therefore, we use the regular structure of the DGG to show that the limiting eigenvalue distribution of the RGG normalized Laplacian matrix converges to a distribution with a Dirac atomic measure at zero. In the thermodynamic regime, we approximate the eigenvalues of the regularized normalized Laplacian matrix of the RGG by the eigenvalues of the DGG regularized normalized Laplacian and we provide an error bound which is valid w.h.p. and depends upon the average vertex degree.

This work presents a new way of exploiting non-uniform file popularity in coded caching networks. Focusing on a fully-connected fully-interfering wireless setting with multiple cache-enabled transmitters and receivers, we show how non-uniform file popularity can be used very efficiently to accelerate the impact of transmitter-side data redundancy on receiver-side coded caching. This approach is motivated by the recent discovery that, under any realistic file-size constraint, having content appear in multiple transmitters can in fact dramatically boost the speed-up factor attributed to coded caching. We formulate an optimization problem that exploits file popularity to optimize the placement of files at the transmitters. We then provide a proof that reduces significantly the variable search space, and propose a new search algorithm that solves the problem at hand. We also prove an analytical performance upper bound, which is in fact met by our algorithm in the regime of many receivers. Our work reflects the benefits of allocating higher cache redundancy to more popular files, but also reflects a law of diminishing returns where for example very popular files may in fact benefit from minimum redundancy. In the end, this work reveals that in the context of coded caching, employing multiple transmitters can be a catalyst in fully exploiting file popularity, as it avoids various asymmetry complications that appear when file popularity is used to alter the receiver-side cache placement., 14 pages, submitted to Transactions on Networking

International audience; In this work the RF propagation has been characterized first for a 5G Wireless Train Backbone, based on a parametric simulation model of a train. Different parameters have been changed to study their effect on the RF propagation. According to these results, some recommendations are made. The antenna propagation has also been studied in a model for the Wireless Consist Network.

International audience; The analysis of memory dumps presents unique challenges, as operating systems use a variety of (often undocumented) ways to represent data in memory. To solve this problem, forensics tools maintain collections of models that precisely describe the kernel data structures used by a handful of operating systems. However, these models cannot be generalized and developing new models may require a very long and tedious reverse engineering effort for closed source systems. In the last years, the tremendous increase in the number of IoT devices, smart-home appliances and cloud-hosted VMs resulted in a growing number of OSs which are not supported by current forensics tools. The way we have been doing memory forensics until today, based on handwritten models and rules, cannot simply keep pace with this variety of systems. To overcome this problem, in this paper we introduce the new concept of OS-agnostic memory forensics, which is based on techniques that can recover certain forensics information without any knowledge of the internals of the underlying OS. Our approach allows to automatically identify different types of data structures by using only their topological constraints and then supports two modes of investigation. In the first, it allows to traverse the recovered structures by starting from predetermined seeds, i.e., pieces of forensics-relevant information (such as a process name or an IP address) that an analyst knows a priori or that can be easily identified in the dump. Our experiments show that even a single seed can be sufficient to recover the entire list of processes and other important forensics data structures in dumps obtained from 14 different OSs, without any knowledge of the underlying kernels. In the second mode of operation, our system requires no seed but instead uses a set of heuristics to rank all memory data structures and present to the analysts only the most 'promising' ones. Even in this case, our experiments show that an analyst can use our approach to easily identify forensics-relevant structured information in a truly OS-agnostic scenario.

International audience; We here continue the exploration of vector coded caching and its powerful effect on the performance of multiuser multiple-input multiple-output (MU-MIMO) systems. In particular, this paper investigates the impact that vector coded caching under various practical considerations, which include channel state information acquisition costs, multi-antenna receivers, variable pathloss, as well as max-min fairness where the minimum transmission rate among the simultaneously served users is maximized via power allocation. We here focus on the widely adopted zero-forcing precoder, and proceed to show the large multiplicative boost brought about by vector coded caching over the optimized cacheless counterpart. To do so, we here provide lower and upper bounds on the overall throughput as well as on the effective gain (i.e., on the rate improvement over the optimized cacheless counterpart), again in the presence of the aforementioned practical aspects. Finally various numerical results demonstrate the tightness of the derived upper and lower bounds, as well as nicely illustrate the notable gains.

International audience; A science is defined by a set of encyclopedic knowledge related to facts or phenomena following rules or evidenced by experimentally-driven observations. Computer Science and in particular computer networks is a relatively new scientific domain maturing over years and adopting the best practices inherited from more fundamental disciplines. The design of past, present and future networking components and architectures have been assisted, among other methods, by experimentally-driven research and in particular by the deployment of test platforms, usually named as testbeds. However, often experimentally-driven networking research used scattered methodologies, based on ad-hoc, small-sized testbeds, producing hardly repeatable results. We believe that computer networks needs to adopt a more structured methodology, supported by appropriate instruments, to produce credible experimental results supporting radical and incremental innovations. This paper reports lessons learned from the design and operation of test platforms for the scientific community dealing with digital infrastructures. We introduce the SLICES initiative as the outcome of several years of evolution of the concept of a networking test platform transformed into a scientific instrument. We address the challenges, requirements and opportunities that our community is facing to manage the full research-life cycle necessary to support a scientific methodology.

This work studies a well-known shared-cache coded caching scenario where each cache can serve an arbitrary number of users, analyzing the case where there is some knowledge about such number of users (i.e., the topology) during the content placement phase. Under the assumption of regular placement and a cumulative cache size that can be optimized across the different caches, we derive the fundamental limits of performance by introducing a novel cache-size optimization and placement scheme and a novel information-theoretic converse. The converse employs new index coding techniques to bypass traditional uniformity requirements, thus finely capturing the heterogeneity of the problem, and it provides a new approach to handle asymmetric settings. The new fundamental limits reveal that heterogeneous topologies can in fact outperform their homogeneous counterparts where each cache is associated to an equal number of users. These results are extended to capture the scenario of topological uncertainty where the perceived/estimated topology does not match the true network topology. This scenario is further elevated to the stochastic setting where the user-to-cache association is random and unknown, and it is shown that the proposed scheme is robust to such noisy or inexact knowledge on the topology., 39 pages

—Recommendations are employed by Content Providers (CPs) of streaming services in order to boost user engagement and their revenues. Recent works suggest that nudging recommendations towards cached items can reduce operational costs in the caching networks, e.g., Content Delivery Networks (CDNs) or edge cache providers in future wireless networks. However, cache-friendly recommendations could deviate from users’ tastes, and potentially affect the CP’s revenues. Motivated by real-world business models, this work identifies the misalignment of the financial goals of the CP and the caching network provider, and presents a network-economic framework for recommendations. We propose a cooperation mechanism leveraging the Nash bargaining solution that allows the two entities to jointly design the recommendation policy. We consider different problem instances that vary on the extent these entities are willing to share their cost and revenue models, and propose two cooperative policies, CCR and DCR, that allow them to make decisions in a centralized or distributed way. In both cases, our solution guarantees reaching a fair and Pareto optimal allocation of the cooperation gains. Moreover, we discuss the extension of our framework towards caching decisions. A wealth of numerical experiments in realistic scenarios show the policies lead to significant gains for both entities.

International audience; In this paper, we unveil the Software-Defined Low Latency (SDLL) framework based on Software-Defined Networking (SDN) to provision the Quality of Service (QoS) and guarantee ultra-low latency in 5G and beyond Transport Networks (TN). SDLL aims to tackle Time Sensitive Networking (TSN)’s weaknesses by providing agility and flexibility in terms of Traffic Engineering (TE) and Queue Management (QM) to guarantee low end-to-end (E2E) latency even under congested links. SDLL provides a flexible and on-demand way to change end-to-end paths and queue configurations (ex., add or remove queues). We conducted extensive experimentation by implementing SDLL using Open Network Operating System (ONOS) and Open vSwitch (OVS) tools and comparing its performances against two standard solutions: SDN Shortest Path (SDNSP) (one queue per port) and Software-Defined QoS (SDQoS) (three queues per port). Obtained results indicate that SDLL can guarantee low E2E latency compared to the two other solutions, particularly when: (1) the links are congested and (2) many low-latency critical services are run in parallel.

The multi-user linearly-separable distributed computing problem is considered here, in which $N$ servers help to compute the real-valued functions requested by $K$ users, where each function can be written as a linear combination of up to $L$ (generally non-linear) subfunctions. Each server computes a fraction $\gamma$ of the subfunctions, then communicates a function of its computed outputs to some of the users, and then each user collects its received data to recover its desired function. Our goal is to bound the ratio between the computation workload done by all servers over the number of datasets. To this end, we here reformulate the real-valued distributed computing problem into a matrix factorization problem and then into a basic sparse recovery problem, where sparsity implies computational savings. Building on this, we first give a simple probabilistic scheme for subfunction assignment, which allows us to upper bound the optimal normalized computation cost as $\gamma \leq \frac{K}{N}$ that a generally intractable $\ell_0$-minimization would give. To bypass the intractability of such optimal scheme, we show that if these optimal schemes enjoy $\gamma \leq - r\frac{K}{N}W^{-1}_{-1}(- \frac{2K}{e N r} )$ (where $W_{-1}(\cdot)$ is the Lambert function and $r$ calibrates the communication between servers and users), then they can actually be derived using a tractable Basis Pursuit $\ell_1$-minimization. This newly-revealed connection between distributed computation and compressed sensing opens up the possibility of designing practical distributed computing algorithms by employing tools and methods from compressed sensing., Comment: Submitted to ITW2023. arXiv admin note: text overlap with arXiv:2206.11119

We consider the problem of distributed lossless computation of a function of two sources by one common user. To do so, we first build a bipartite graph, where two disjoint parts denote the individual source outcomes. We then project the bipartite graph onto each source to obtain an edge-weighted characteristic graph (EWCG), where edge weights capture the function's structure, by how much the source outcomes are to be distinguished, generalizing the classical notion of characteristic graphs. Via exploiting the notions of characteristic graphs, the fractional coloring of such graphs, and edge weights, the sources separately build multi-fold graphs that capture vector-valued source sequences, determine vertex colorings for such graphs, encode these colorings, and send them to the user that performs minimum-entropy decoding on its received information to recover the desired function in an asymptotically lossless manner. For the proposed EWCG compression setup, we characterize the fundamental limits of distributed compression, verify the communication complexity through an example, contrast it with traditional coloring schemes, and demonstrate that we can attain compression gains higher than $\% 30$ over traditional coloring., Comment: appeared in IEEE ISIT 2023

International audience; Delivering ultra-reliable low-latency communication (URLLC) services on 5G New Radio (NR) is critical for extended Vehicle-to-everything (V2X) applications. Early studies focused either on infrastructure V2X communications, or exclusively evaluating performance on mmWave band, but never so far for V2X Sidelink communications. In this paper, we evaluate the feasibility of achieving URLLC critical requirements for 5G NR V2X Sidelink communication at the V2X 5.9 GHz band. We propose a higher 5G-NR numerology completed by a deterministic scheduler and propose an analytical V2X SL admission control algorithm to reach tight URLLC requirements.

With its low latency and high speeds, 5G aims to support vertical services like automotive, industry, agriculture, and manufacturing. Unlike the precedent mobile generations, which just provided voice and data to domestic and business customers, 5G aims to create a common infrastructure to support diverse requirements of the vertical industry's needs in terms of communication and networking capabilities, such as high data rate and low latency. But before a commercial deployment, a trial phase is needed to validate that the network can support these requirements. In this context, several 5G facilities have been established to run 5G trials, mainly built to be used by networking experts. However, vertical service owners have small to no knowledge of the technical details of the 5G infrastructure. In this paper, we introduce EURECOM 5G facility, which was specifically designed to run vertical use cases by abstracting and simplifying as much as possible the trial deployment and Key Performance Indicator (KPI) collection. EURECOM 5G facility provides a rich number of 5G components to test, including 5G New Radio, Network Slicing, Edge Computing, and KPI visualization, allowing verticals to have a real 5G environment for testing their applications and services. Finally, the facility relies mainly on open-source components.

Evaluating the challenges and opportunities of cooperative autonomous vehicles (CAV) require an adapted simulation methodology reproducing realistic driving and sensory contexts. In this paper, we propose a RounD-like CARLA scenario reproducing in CARLA the driving context recorded in the RounD dataset. We focus in particular on roundabout scenarios, as they are considered particularly challenging for CAV. We present the methodology followed to generate the CARLA scenario and describe challenges to reproduce trajectories corresponding to RounD. Origin and destination of vehicles, waypoint and speed are extracted from RounD for CARLA vehicles to closely reproduce the driving patterns observed in RounD.The benefit of such scenario are manyfold, such as evaluating control algorithms of CAVs, deep AI reinforcement learning, or vehicular sensor data sampling under realistic driving contexts. It notably will reduce the gap of AI mechanisms for CAV between simulation scenarios and realistic conditions.