Search

Vehicular social network is emerging as a new promising concept, combining two types of network paradigms, namely, vehicular networks and social networks. In order to manage efficiently th...

Preprint submitted to EJOR; We consider a telecommunications company expanding its network capacity to face an increasing demand. The company can also invest in marketing to incentivize clients to shift to more recent technologies, hopefully leading to cheaper overall costs. To model the effect of marketing campaigns, previous works have relied on the Bass model. Since that model only provides a rough approximation of the actual shifting mechanism, the purpose of this work is to consider uncertainty in the shifting mechanism through the lens of robust optimization. We thus assume that the (discrete) shifting function can take any value in a given polytope and wish to optimize against the worst-case realization. The resulting robust optimization problem possesses integer recourse variables and non-linear dependencies on the uncertain parameters. We address these difficulties as follows. First, the integer recourse is tackled heuristically through a piece-wise constant policy dictated by a prior partition of the uncertainty polytope. Second, the non-linearities are handled by a careful analysis of the dominating scenarios. The scalability and economical relevance of our models are assessed through numerical experiments performed on realistic instances. In particular, we choose one of these instances to perform a case study with simulations illustrating the possible benefit of using robust optimization.

L'objectif principal de ce livre consiste à étudier les solutions susceptibles d'assurer la sécurité dans les réseaux mobiles ad hoc, en proposant une architecture hiérarchique distribuée qui permet d'établir une infrastructure dynamique à clé publique. Cette architecture doit supporter les différentes caractéristiques de ces réseaux. Dans ce but, un modèle de confiance adapté à l'environnement dynamique pour assurer l'évolution des niveaux de confiance des nœuds est établi. Dans le but de sécuriser les nœuds dont le rôle est crucial au sein du réseau, leur identité doit être cachée. C'est pourquoi le concept d'anonymat est introduit. Un protocole d'authentification anonyme est proposé. De plus, pour entretenir le modèle de confiance, un mécanisme de surveillance est indispensable. Il est adapté aux contraintes de l'environnement sans fil dynamique et réduit le taux de fausses alarmes. Il est fondé sur une approche inter-couches et un modèle probabiliste pour améliorer l'observation du nœud surveillant. Pour faire face aux attaques intelligentes de type inter-couches, une étude des vulnérabilités au niveau des couches inférieures comme la couche MAC est menée.

International audience; In this paper, we propose a new architecture based on an efficient trust model and Secure Distributed Clustering Algorithm (SDCA) in order to distribute a certification authority (CA) for ensuring the distribution of certificates in each cluster. We use the combination of a fully self-organized security for trust models like PGP adapted to ad-hoc technology and the clustering algorithm which is based on the use of trust and mobility metrics, in order to select the clusterhead and to establish PKI in each cluster for authentication and exchange of data. Furthermore, we present a new approach: the Dynamic Demilitarized Zone (DDMZ) to protect the CA in each cluster. The principal idea of DDMZ consists in selecting the dispensable nodes, also called registration authorities (RA); these nodes must be confident and located at one-hope from the CA. Their roles are to receive, filter and treat the requests from any unknown node to the CA. With this approach, we can avoid the single point of failure in each cluster. Moreover, we propose a probabilistic model to define the direct connectivity between confident nodes in order to study the resistance degree of the DDMZ against different attacks. In addition, we evaluate the performance of the proposed SDCA and we estimate the robustness and the availability of DDMZ through the simulations. The effects of direct connectivity and transmission range on the stability and security of the network are analysed. The simulation's results confirm that the proposed architecture is scalable, secure and more resistant against attacks.

International audience; To avoid the single point of failure for the certificate authority (CA) in MANET, a decentralized solution is proposed where nodes are grouped into different clusters. Each cluster should contain at least two confident nodes. One is known as CA and the another as register authority RA. The Dynamic Demilitarized Zone (DDMZ) is proposed as a solution for protecting the CA node against potential attacks. It is formed from one or more RA node. The problems of such a model are: (1) Clusters with one confident node, CA, cannot be created and thus clusters' sizes are increased which negatively affect clusters' services and stability. (2) Clusters with high density of RA can cause channel collision at the CA. (3) Clusters' lifetime are reduced since RA monitors are always launched (i.e., resource consumption). In this paper, we propose a model based on mechanism design that will allow clusters with single trusted node (CA) to be created. Our mechanism will motivate nodes that do not belong to the confident community to participate by giving them incentives in the form of trust, which can be used for cluster's services. To achieve this goal, a RA selection algorithm is proposed that selects nodes based on a predefined selection criteria function and location (i.e., using directional antenna). Such a model is known as moderate. Based on the security risk, more RA nodes must be added to formalize a robust DDMZ. Here, we consider the tradeoff between security and resource consumption by formulating the problem as a nonzero-sum noncooperative game between the CA and attacker. Finally, empirical results are provided to support our solutions.

International audience; The intrusion detection system (IDS) for mobile ad hoc networks (MANET) consists in monitoring the nodes' behavior, in order to detect the malicious activity of nodes. Many existing solutions deal with the problem at each layer separately. But new kinds of misbehavior attacks are cross-layer attacks. And such smart misbehaviors cannot be detected at the level of one layer. In this paper, we propose a new cross-layer approach based on physical, MAC, and routing layers for a monitoring mechanism. A new analytical model is proposed to illustrate the parameters' effect on these different layers. The impact of the signal to noise ratio (SNR) and the distance between monitor and monitored nodes are clearly introduced. Moreover, the difference between the carrier sense, the interference range, and the transmission range is taken into account in our model. The proposed model improves the evaluation of the nodes' cooperation and reduces the risk of having any false positive rate. The analytical study and simulation results illustrate our purpose. In addition, with the simulations' results, we illustrate the impact of the distance between monitor and monitored nodes on the monitoring mechanism. Finally, we show that our cross-layer mechanism has a lower false positive rate than the classical Watchdog mechanism in different network's parameters such as the nodes' density, the speed mobility, and the different traffic loads.

International audience; In this paper, we focus on the Medium Access Control (MAC), particularly the IEEE 802.11 and we deal with some hidden vulnerabilities based on the control packets CTS (Clear to Send) and ACK. Through these vulnerabilities, we show two new smart attacks which were not dealt by the solutions proposed recently like the attack based on the RTS (Request to Send) packet vulnerability. The malicious node can exploit these vulnerabilities on the MAC protocol, in order to corrupt the monitoring and routing processes. Furthermore, we demonstrate the attacks through algorithms and we show how vulnerabilities can be exploited and how these attacks can be implemented by the attacker. The impact of these attacks is presented through simulation and implementation. Simulation and exprimental results show the impact of the attacks on the network. In addition, the experimental results demonstrate the feasibility of these real attacks and their exploitation. These experimentations allow us to confirm the simulation's results. Furthermore, in order to prevent these attacks, the solutions based on control packet authentication are presented. We propose two kinds of the solution one is cryptography independent and the other one is cryptography dependent. The evaluation and analysis of these solutions are investigated by analytic and simulations analysis. The simulations' results of the proposed solution show that the attacks are prevented and the negative impacts are significantly reduced. In addition, the security cost of the proposed solutions are investigated. Hence, the security costs are insignificant in comparison with the negative impact of these attacks.

International audience; In this paper, we propose a new architecture based on an efficient trust model and clustering algorithm in order to distribute a certification authority (CA) for ensuring the distribution of certificates in each cluster. We use the combination of fully self-organized security for trust model like PGP adapted to ad-hoc technology and the clustering algorithm which is based on the use of trust and mobility metric, in order to select the clusterhead and to establish PKI in each cluster for authentication and exchange of data. Furthermore, we present new approach Dynamic Demilitarized Zone (DDMZ) to protect CA in each cluster. The principle idea of DDMZ consists to select the dispensable nodes, also called registration authorities; these nodes must be confident and located at one-hope from the CA. Their roles are to receive, filter and treat the requests from any unknown node to CA. With this approach, we can avoid the single point of failure in each cluster. This architecture can be easily extended to other hierarchical routing protocols. Simulation results confirm that our architecture is scalable and secure.

International audience; We investigate algorithms that solve exactly the robust single machine scheduling problem that minimizes the total tardiness. We model the processing times as uncertain and let them take any value in a budgeted uncertainty set. Therefore , the objective seeks to minimize the worst-case tardiness over all possible values. We compare, through computational experiments, two types of solution algorithms. The first combines classical MILP formulations with row-and-column generation algorithms. The second generalizes the classical branch-and-bound algorithms to the robust context, extending state-of-the-art concepts used for the deterministic version of the problem. By generalizing the classical branch-and-bound algorithm we are able to assemble and discuss good algorithmic decisions steps that once put together make our robust branch-and-bound case attractive. For example, we extend and adapt dominance rules to our uncertain problem, making them an important component of our robust algorithms. We assess our algorithms on instances inspired by the scientific literature and identify under what conditions an algorithm has better performance than others. We introduce a new classifying parameter to group our instances, also extending existing methods for the deterministic problem case.

International audience; We study the scenario where individuals (speakers) contribute to the publication of an anonymized speech corpus. Data users then leverage this public corpus to perform downstream tasks (such as training automatic speech recognition systems), while attackers may try to de-anonymize itbased on auxiliary knowledge they collect. Motivated by this scenario, speaker anonymization aims to conceal the speaker identity while preserving the quality and usefulness of speech data. In this paper, we study x-vector based speaker anonymization, the leading approach in the recent Voice Privacy Challenge, which converts an input utterance into that of a random pseudo-speaker. We show that the strength of the anonymization varies significantly depending on how the pseudo-speaker is selected. In particular, we investigate four design choices: the distance measure between speakers, the region of x-vector space where the pseudo-speaker is mapped, the gender selection and whether to use speaker or utterance level assignment. We assess the quality of anonymization from the perspective of the three actors involved in our threat model, namely the speaker, the user and the attacker. To measure privacy and utility, we use respectively the linkability score achieved by the attackers and the decoding word error rate incurred by an ASR model trained with the anonymized data. Experiments on LibriSpeech dataset confirm that the optimal combination ofdesign choices yield state-of-the-art performance in terms of privacy protection as well as utility. Experiments on Mozilla Common Voice dataset show that the best design choices with 50 speakers guarantee the same anonymization level against re-identification attack as raw speech with 20,000 speakers.

International audience

International audience

Supplementary material to the paper "The VoicePrivacy 2020 Challenge: Results and findings" (https://hal.archives-ouvertes.fr/hal-03332224) submitted to CSL.; The VoicePrivacy 2020 Challenge focuses on developing anonymization solutions for speech technology. This report complements the summary results and analyses presented by Tomashenko et al. (2021). After quickly recalling the challenge design and the submitted anonymization systems, we provide more detailed results and analyses. First, we present objective evaluation results for the primary challenge metrics and for alternative metrics and attack models, and we compare them with each other. Second, we present subjective evaluation results for speaker verifiability, speech naturalness, and speech intelligibility. Finally, we compare these objective and subjective evaluation results with each other.

International audience; Multi-label text classification consists in attributing, for each textual document, one or more labels. Due to its nature, the task is often considered to be more challenging than other types of classification problems since the number of labels to assign is unknown. In text documents, this difficulty is generally the result of a blurry border between lexical fields of the labels or an underrepresentation of some of them. In this paper, we seek to automatically associate categories to scientific articles related to the COVID-19. We propose to address this multi-label classification problem by integrating an original keyword enhancement method to the TARS transformer-based approach designed to perform few-shot learning. Experiments conducted during the BioCreative challenge on the multi-label classification task show that our approach outperforms the baseline (ML-Net), no matter the metric considered.

This paper presents the results and analyses stemming from the first VoicePrivacy 2020 Challenge which focuses on developing anonymization solutions for speech technology. We provide a systematic overview of the challenge design with an analysis of submitted systems and evaluation results. In particular, we describe the voice anonymization task and datasets used for system development and evaluation. Also, we present different attack models and the associated objective and subjective evaluation metrics. We introduce two anonymization baselines and provide a summary description of the anonymization systems developed by the challenge participants. We report objective and subjective evaluation results for baseline and submitted systems. In addition, we present experimental results for alternative privacy metrics and attack models developed as a part of the post-evaluation analysis. Finally, we summarize our insights and observations that will influence the design of the next VoicePrivacy challenge edition and some directions for future voice anonymization research., Comment: Submitted to the Special Issue on Voice Privacy (Computer Speech and Language Journal - Elsevier); under review

Attribute-driven privacy aims to conceal a single user's attribute, contrary to anonymisation that tries to hide the full identity of the user in some data. When the attribute to protect from malicious inferences is binary, perfect privacy requires the log-likelihood-ratio to be zero resulting in no strength-of-evidence. This work presents an approach based on normalizing flow that maps a feature vector into a latent space where the evidence, related to the binary attribute, and an independent residual are disentangled. It can be seen as a non-linear discriminant analysis where the mapping is invertible allowing generation by mapping the latent variable back to the original space. This framework allows to manipulate the log-likelihood-ratio of the data and therefore allows to set it to zero for privacy. We show the applicability of the approach on an attribute-driven privacy task where the sex information is removed from speaker embeddings. Results on VoxCeleb2 dataset show the efficiency of the method that outperforms in terms of privacy and utility our previous experiments based on adversarial disentanglement., ICASSP 2022

In presence of a high traffic load, not all offered traffic can be satisfied at all time. Luckily, several kinds of services (e.g. video, file transfer) are elastic because they can be degraded and accepted with a lower bit-rate, corresponding to a lower quality of service (QoS) level. In this paper, we aim at maximizing network revenue leveraging elastic service provisioning. To tackle this problem, we propose an integer linear programming model as well as a decomposition method, which selects a QoS level and provisions a lightpath for elastic request. Meanwhile, in order to exploit the spectrum resources of elastic services under a progressive network load, we propose an auto-degrading provisioning scheme to trigger the network reconfiguration for existing lightpaths in an effort to increase the total network revenue. Simulation results validate the revenue improvement by supporting elastic service provisioning scheme in the scenarios of static and progressive network load.

International audience; Our study examined the performance of evaluators tasked to group natural and anonymised speech recordings into clusters based on their perceived similarities. Speech stimuli were selected from the VCTK corpus; two systems developed for the VoicePrivacy 2020 Challenge were used for anonymisation. The Baseline-1 (B1) system was developed by using x-vectors and neural waveform models, while the Baseline-2 (B2) system relied on digital-signal-processing techniques. 74 evaluators completed three trials composed of 16 recordings with either natural or anonymised speech generated from a single system. F-measure and cluster purity metrics were used to assess evaluator accuracy. Probabilistic linear discriminant analysis (PLDA) scores from an automatic speaker verification system were generated to quantify similarity between recordings and used to correlate subjective results. Our findings showed that non-native English speaking evaluators significantly lowered their F-measure means when presented anonymised recordings. We observed no significance for cluster purity. Pearson correlation procedures revealed that PLDA scores generated from natural and B2-anonymised speech recordings correlated positively to F-measure and cluster purity metrics. These findings show evaluators were able to use the interface to cluster natural and anonymised speech recordings and suggest anonymisation systems modelled like B1 are more effective at suppressing identifiable speech characteristics.

Deep learning contributes to reaching higher levels of artificial intelligence. Due to its pervasive adoption, however, growing concerns on the environmental impact of this technology have been raised. In particular, the energy consumed at training and inference time by modern neural networks is far from being negligible and will increase even further due to the deployment of ever larger models. This work investigates for the first time the carbon cost of end-to-end automatic speech recognition (ASR). First, it quantifies the amount of CO2 emitted while training state-of-the-art (SOTA) ASR systems on a university-scale cluster. Then, it shows that a tiny performance improvement comes at an extremely high carbon cost. For instance, the conducted experiments reveal that a SOTA Transformer emits 50% of its total training released CO2 solely to achieve a final decrease of 0.3 of the word error rate. With this study, we hope to raise awareness on this crucial topic and we provide guidelines, insights, and estimates enabling researchers to better assess the environmental impact of training speech technologies.

International audience; This paper describes the ON-TRAC Consortium translation systems developed for two challenge tracks featured in the Evaluation Campaign of IWSLT 2021, low-resource speech translation and multilingual speech translation. The ON-TRAC Consortium is composed of researchers from three French academic laboratories and an industrial partner: LIA (Avignon Université), LIG (Université Grenoble Alpes), LIUM (Le Mans Université), and researchers from Airbus. A pipeline approach was explored for the lowresource speech translation task, using a hybrid HMM/TDNN automatic speech recognition system fed by wav2vec features, coupled to an NMT system. For the multilingual speech translation task, we investigated the use of a dual-decoder Transformer that jointly transcribes and translates an input speech signal. This model was trained in order to translate from multiple source languages to multiple target ones.

International audience; This article is a contribution to the analysis of the effects of social networks on the shaping of taste and opinions concerning cultural matters. Taking as an example the Twitter usage of thirty-two professional partners of Vodkaster who play with each other on the Cannes stage and using mainly lexical ana­lysis, one can shed light on three structural concepts: the digital outsider – his deviance being a central feature of the event, lea­dership – based on faithful readership, and the fourth circle of spectators who experience the festival by proxy. Thus, digital social networks are both an easily mobilized resource for the analysis of audience and a specific form of festival-goer’s sociability.; Este artículo es una contribución al análisis de los efectos de las redes sociales sobre la formación de los gus­tos y de las opiniones en materia cultural. Tomando el ejemplo de los usos de Twitter en el festival de Cannes de treinta y dos profesionales (socios de Vodkaster) que tocan juntos en la escena local, y apoyados fundamentalmente en un análisis lexical, pode­mos dar a luz tres nociones estructurantes : la de outsider numérico (la desviación constitutiva del acontecimiento), la de liderazgo fundada sobre la fidelidad de los lectores y la del cuarto círculo de espectadores que construye su experiencia del festival en su nombre. Las redes sociales constituyen a la vez un recurso fácilmente explotable para el análisis de los públicos y una forma específica de la sociabilidad del festival.; Cet article est une contribution à l’analyse des effets des réseaux sociaux sur la formation des goûts et des opinions en matière culturelle. En prenant l’exemple des usages de Twitter au Festival de Cannes par trente-deux professionnels partenaires de Vodkaster qui jouent ensemble sur la scène can­noise et en s’appuyant principalement sur une analyse lexicale, on peut mettre au jour trois notions structurantes : celle d’outsi­der numérique – la déviance étant constitutive de l’événement –, celle de leadership fondée sur la fidélité du lectorat, et celle d’un quatrième cercle de spectateurs qui construit son expérience du Festival par procuration. Les réseaux sociaux numériques consti­tuent ainsi à la fois une ressource aisément exploitable pour l’ana­lyse des publics et une forme spécifique de la sociabilité festivalière.

In this report, the fundamental limits on the rates at which information and energy can be simultaneously transmitted over an additive white Gaussian noise channel are studied under the following assumptions: (a) the channel is memoryless; (b) the number of channel input symbols (constellation size) and block length are finite; and (c) the decoding error probability (DEP) and the energy outage probability (EOP) are bounded away from zero. In particular, it is shown that the limits on the maximum information and energy transmission rates; and the minimum DEP and EOP, are essentially set by the type induced by the code used to perform the transmission. That is, the empirical frequency with which each channel input symbol appears in the codewords. Using this observation, guidelines for optimal constellation design for simultaneous energy and information transmission are presented.; Dans ce rapport, les limites fondamentales des débits auxquelles l’information et l’énergie peuvent être transmises simultanément sur un canal de bruit gaussien blanc additif sont étudiées sous les hypothèses suivantes : (a) le canal est sans mémoire; (b) le nombre de symboles d’entrée du canal (taille de la constellation) et la durée de la transmission sont finis ; et (c) la probabilité d’erreur de décodage (DEP) et la probabilité de coupure d’énergie (EOP) sont limitées par des bornes inférieures strictement positives. En particulier, il est montré que les limites sur les débits maximaux de transmission d’information et d’énergie et les DEP et EOP minimaux, sont essentiellement fixés par le type induit par le code utilisé pour effectuer la transmission. C’est-à-dire la fréquence empirique à laquelle chaque symbole d’entrée de canal apparaît dans les mots de code. En utilisant cette observation, des recommandations pour la conception optimale des constellations pour la transmission simultanée dinformation et d’énergie sont présentées.

Self-Supervised Learning (SSL) using huge unlabeled data has been successfully explored for image and natural language processing. Recent works also investigated SSL from speech. They were notably successful to improve performance on downstream tasks such as automatic speech recognition (ASR). While these works suggest it is possible to reduce dependence on labeled data for building efficient speech systems, their evaluation was mostly made on ASR and using multiple and heterogeneous experimental settings (most of them for English). This questions the objective comparison of SSL approaches and the evaluation of their impact on building speech systems. In this paper, we propose LeBenchmark: a reproducible framework for assessing SSL from speech. It not only includes ASR (high and low resource) tasks but also spoken language understanding, speech translation and emotion recognition. We also focus on speech technologies in a language different than English: French. SSL models of different sizes are trained from carefully sourced and documented datasets. Experiments show that SSL is beneficial for most but not all tasks which confirms the need for exhaustive and reliable benchmarks to evaluate its real impact. LeBenchmark is shared with the scientific community for reproducible research in SSL from speech., Will be presented at Interspeech 2021

International audience; We consider a network with Massive Multiple Input Multiple Output (M-MIMO) base stations using a Grid of Beams (GoB) for data and control channels. 5G allows to establish interference relations between beams of neighboring cells. Such relations can be used to automatically generate a beam relation matrix, denoted as Automatic Neighbor Beam Relation (ANBR) matrix that can be very useful for optimizing different resource allocation processes. This paper shows how the ANBR matrix can be used to coordinate scheduling of neighboring cells with a small amount of information exchange. The coordination is performed by judicially muting or reducing the band-width of certain beams in the process of MultiUser (MU) Proportional Fair (PF) scheduling. Numerical results show how the coordination approach can bring about significant performance gain.

Pre-training for feature extraction is an increasingly studied approach to get better continuous representations of audio and text content. In the present work, we use wav2vec and camemBERT as self-supervised learned models to represent our data in order to perform continuous emotion recognition from speech (SER) on AlloSat, a large French emotional database describing the satisfaction dimension, and on the state of the art corpus SEWA focusing on valence, arousal and liking dimensions. To the authors' knowledge, this paper presents the first study showing that the joint use of wav2vec and BERT-like pre-trained features is very relevant to deal with continuous SER task, usually characterized by a small amount of labeled training data. Evaluated by the well-known concordance correlation coefficient (CCC), our experiments show that we can reach a CCC value of 0.825 instead of 0.592 when using MFCC in conjunction with word2vec word embedding on the AlloSat dataset., Comment: Accepted in IEEE SLT 2021

International audience; Background: Speech disorders impact quality of life for patients treated with oral cavity and oropharynx cancers. However, there is a lack of uniform and applicable methods for measuring the impact on speech production after treatment in this tumor location. Objective: The objective of this work is to (1) model an automatic severity index of speech applicable in clinical practice, that is equivalent or superior to a severity score obtained by human listeners, via several acoustics parameters extracted (a) directly from speech signal and (b) resulting from speech processing and (2) derive an automatic speech intelligibility classification (i.e., mild, moderate, severe) to predict speech disability and handicap by combining the listener comprehension score with self-reported quality of life related to speech. Methods: Eighty-seven patients treated for cancer of the oral cavity or the oropharynx and 35 controls performed different tasks of speech production and completed questionnaires on speech-related quality of life. The audio recordings were then evaluated by human perception and automatic speech processing. Then, a score was developed through a classic logistic regression model allowing description of the severity of patients' speech disorders. Results: Among the group of parameters subject to extraction from automatic processing of the speech signal, six were retained, producing a correlation at 0.87 with the perceptual reference score, 0.77 with the comprehension score, and 0.5 with speech-related quality of life. The parameters that contributed the most are based on automatic speech recognition systems. These are mainly the automatic average normalized likelihood score on a text reading task and the score of cumulative rankings on pseudowords. The reduced automatic YC2SI is modeled in this way: YC2SIp = 11.48726 + (1.52926 × Xaveraged normalized likelihood reading) + (−1.94e-06 × Xscore of cumulative ranks pseudowords). Conclusion: Automatic processing of speech makes it possible to arrive at valid, reliable, and reproducible parameters able to serve as references in the framework of follow-up of patients treated for cancer of the oral cavity or the oropharynx.

International audience; Network Slicing is one of the essential concepts that has been introduced in 5G networks design to support demand expressed by next generation services. Network slicing will also bring new business opportunities for service providers (SPs) and virtual network operators, allowing them to run their virtual, independent business operations on shared physical infrastructure. We consider a marketplace where service providers (SPs) i.e., slice tenants lease the resources from an infrastructure provider (InP) through a network slicing mechanism. They compete to offer a certain communication service to end-users. We show that the competition between SPs can be model using the multiresource Tullock contest (TC) framework, where SPs exert effort by expending costly resource to attract users. We study the competition between the SPs under a static and dynamic resource sharing scheme. In a dynamic resource sharing scheme, SPs are pre-assigned with fixed shares (budgets) of infrastructure, and they are allowed to redistribute their shares and customise their allocation to maximise their profit. The decision problem of SPs is analysed using non-cooperative game theory, and it is shown that the resultant game admits a unique Nash Equilibrium (NE). Furthermore, a distributed reinforcement algorithm is proposed that allows each SP to reach the game's unique Nash equilibrium. Finally, simulations results are conducted to analyse the interaction between market players and the economic efficacy of the network sharing mechanism.

In this work, we propose a Graph Convolutional Neural Networks (GCN) based scheduling algorithm for adhoc networks. In particular, we consider a generalized interference model called the $k$-tolerant conflict graph model and design an efficient approximation for the well-known Max-Weight scheduling algorithm. A notable feature of this work is that the proposed method do not require labelled data set (NP-hard to compute) for training the neural network. Instead, we design a loss function that utilises the existing greedy approaches and trains a GCN that improves the performance of greedy approaches. Our extensive numerical experiments illustrate that using our GCN approach, we can significantly ($4$-$20$ percent) improve the performance of the conventional greedy approach., Comment: 10 pages, Accepted at EAI VALUETOOLS 2021 - 14th EAI International Conference on Performance Evaluation Methodologies and Tools

International audience

Cet article présente la collecte et l’analyse — à l’aide les logiciels Modalisa et Tropes — de 1867 commentaires d’abonnés d’un corpus de 10 vidéos pour « débutants ou grands débutants » à la guitare. Il s’appuie sur deux entretiens semi-directifs avec le créateur de la chaîne MrGalagomusic, premier vidéaste du web français en nombre d’abonnés qui propose des tutoriels d’apprentissage de la guitare. Il s’agit de comprendre comment la forme « tutoriel » permet le partage d’expériences, de savoirs, de compétences qui ne relèvent pas du schéma classique et linéaire de la transmission en Sciences de l’information et de la communication. Le « tuto » sur YouTube engage une communication du youtubeur vers l’internaute, de l’internaute vers le youtubeur ou de l’internaute vers un autre internaute par la possibilité de laisser un commentaire et d’y répondre. Il s’apparente davantage aux formes de communication orchestrale qui reposent sur l’interaction et constitue un objet pertinent pour comprendre les logiques de partage, d’échange, de contribution et de participation engagées sur les interfaces numériques et utile pour comprendre les nouveaux enjeux qui se fixent dans les dynamiques liées aux questions d’éducation artistique et culturelle., This article presents the collection and analysis - via the Modalisa and Tropes softwares - of 1867 comments from subscribers to 10 videos for "beginners or great beginners" on guitar. It relies on two semi-structured interviews with the creator of the chain MrGalagomusic, the first videographer of the French web in number of subscribers, offering tutorials for learning the guitar. It is to understand how the tutorial form allows the sharing of experiences, knowledge and skills that do not fall within the classical and linear pattern of transmission in the information and communication sciences. The "tutorial" on YouTube commits a communication from youtuber to the user, from the user to the youtuber or from the user to another user by the possibility of leaving a comment and responding. It is more similar to forms of orchestral communication that are based on interaction and is a relevant object to understand the logic of sharing, exchange, contribution and participation engaged on digital interfaces and useful for understanding the new challenges that arise in the dynamics linked to artistic and cultural education issues., Interfaces numériques, 9(2), 2020

International audience; This paper investigates the domain adaptation back-end methods introduced over the past years for speaker recognition, when the mismatch between training and test data induces a severe degradation of performance. The conducted analyses lead to suggest some ways, experimentally validated, for the task of collecting in-domain data and making the most of the first data at hand. The proposed strategy helps to quickly increase accuracy of the detection, without omitting to take into account the practical difficulties of the task of data collecting in real-life situations , and without the cost and delay for forming the expected large and speaker-labeled in-domain dataset. Moreover, a new approach of artificial speaker labeling by clustering is proposed, that dispenses of collecting a preliminary annotated in-domain dataset, with a similar gain of efficiency.

Article soumis et accepté à la conférence Interspeech - Octobre 2020 - Shangai.; International audience; Recent studies have led to the introduction of Speech-to-Concept End-to-End (E2E) neural architectures for Spoken Language Understanding (SLU) that reach state of the art performance. In this work, we propose a way to compute confidence measures on semantic concepts recognized by a Speech-to-Text E2E SLU system. We investigate the use of the hidden representations of our CTC-based SLU system to train an external simple classifier. We experiment two kinds of external simple classifiers to analyze subsequences of hidden representations involved in recognized semantic concepts. The first external classifier is based on a MLP while the second one is based on a bLSTM neural network. We compare them to a baseline confidence measure computed directly from the softmax outputs of the E2E system. On the French challenging MEDIA corpus, when the confidence measure is used to reject, experiments show that using an external BLSTM significantly out-performs the other approaches in terms of precision/recall. To evaluate the additional information provided by this confidence measure, we compute the value of Normalised Cross-Entropy (NCE). Reaching a value equal to 0.288, we show that our best proposed confidence measure brings relevant information about the reliability of a recognized concept.

International audience; Self-supervised learning from raw speech has been proven beneficial to improve automatic speech recognition (ASR). We investigate here its impact on end-to-end automatic speech translation (AST) performance. We use a contrastive predic-tive coding (CPC) model pre-trained from unlabeled speech as a feature extractor for a downstream AST task. We show that self-supervised pre-training is particularly efficient in low resource settings and that fine-tuning CPC models on the AST training data further improves performance. Even in higher resource settings, ensembling AST models trained with filter-bank and CPC representations leads to near state-of-the-art models without using any ASR pre-training. This might be particularly beneficial when one needs to develop a system that translates from speech in a language with poorly standardized orthography or even from speech in an unwritten language. Index Terms: self-supervised learning from speech, automatic speech translation, end-to-end models, low resource settings.

Ce document détaille l’élaboration de la Base de données Révisée des Élu·es de France (BRÉF) à partir d’une source principale, le Répertoire National des Élus (RNE) et de plusieurs sources secondaires, les bases de données de l’Assemblée Nationale, du Sénat et du Parlement Européen. Cette base a vocation a être étendue ultérieurement, en exploitant plus complètement ces sources secondaires, et à plus long terme en intégrant de nouvelles bases de données et des apports ponctuels.

Among metrics that highly affect video quality and quality of experience (QoE), we can cite the delay and cost transmission caused by mobile network overhead. Moreover, Information centric networking (ICN) is a new architecture that is proposed as a technique that offers very high throughput rates and very low latency, especially in QoE sensitive applications such as multimedia content delivery for the future communication networks. Indeed, unused memory in user equipment can be used to cache contents and afford it to the other nearby users on demand. This caching method is considered as one of the most promising solutions to enhance the QoE of users. On the other hand, one of the major research goals is to improve caching nodes decision based on nodes’ characteristics. In this paper, we propose a new programmable architecture named FollowMeCache based on Software Defined Network (SDN), ICN approaches and Fog Computing for cache node selection using the Connected Dominating Sets (CDS) in order to reduce the download delay and cost for the users’ requested video. In fact, we define cache node selection algorithm based on Influence factor to elect the appropriate nodes from the CDS set to build the cache. This metric takes into consideration the connectivity degree, Zone of Interest (ZI), node capacity, user preference and its location. In order to evaluate the proposed solution, we define a smart event use case. The performance evaluation is conducted by using network simulator and the obtained results show that our approach can give significant gain in terms of network throughput and the transmission delay.

This paper studies the improvement in network throughput of an ad hoc network from using both licensed and unlicensed spectra compared to the case where only unlicensed spectrum is used. We address the problem of how the nodes of the network, or secondary users (SUs), should spread their transmissions on both licensed and unlicensed spectra to maximize network throughput, and characterize ‘sharing gain’ achievable in such spectrum sharing systems. The gain obtained can be significant and is increasing with the density of the SUs. The primary and secondary users are modeled as two independent Poisson point processes and their performance is evaluated using techniques from stochastic geometry. A co-operative case is considered where the channel selection strategy of the nodes is centrally controlled. Then, a non-cooperative channel selection game where the SUs selfishly select the channels is analyzed. A pricing scheme is proposed to drive the decisions of SUs to a favorable point. Specifically, by setting an ‘appropriate’ price, global optimal performance is attainable at equilibrium in some cases. Finally, the analysis is extended to the case where a network shares spectrum with a cellular network.

Facebook News Feed personalization algorithm has a significant impact, on a daily basis, on the lifestyle, mood and opinion of millions of Internet users. Nonetheless, the behavior of such algorithm lacks transparency, motivating measurements, modeling and analysis in order to understand and improve its properties. In this paper, we propose a reproducible methodology encompassing measurements, an analytical model and a fairness-based News Feed design. The model leverages the versatility and analytical tractability of time-to-live (TTL) counters to capture the visibility and occupancy of publishers over a News Feed. Measurements are used to parameterize and to validate the expressive power of the proposed model. Then, we conduct a what-if analysis to assess the visibility and occupancy bias incurred by users against a baseline derived from the model. Our results indicate that a significant bias exists and it is more prominent at the top position of the News Feed. In addition, we find that the bias is non-negligible even for users that are deliberately set as neutral with respect to their political views, motivating the proposal of a novel and more transparent fairness-based News Feed design., IFIP Perfomance 2018 - 36th International Symposium on Computer Performance, Modeling, Measurements and Evaluation

International audience; Atypical speech productions, regardless of their origins (accents, learning, pathology), need to be assessed with regard to "typical" or "expected" productions. Evaluation is necessarily based on comparisons between linguistic forms produced and linguistic forms expected. In the field of speech disorders, the intelligibility of a patient is evaluated in order to measure the functional impact of his/her pathology on his/her oral communication. The usual method is to transcribe orthographic linguistic forms perceived and to assign a global and imprecise rating based on their correctness or incorrect. To obtain a more precise evaluation of the production deviations, we propose a measurement method based on phonological transcriptions. An algorithm computes automatically and finely the distances between the phonological forms produced and expected from cost matrices based on the differences of features between phonemes. A first test of this method among a large population of healthy speakers and patients treated for cancer of the oral and pharyngeal cavities has proved its validity.

International audience; We present a new corpus, named AlloSat, composed of real-life call center conversations in French that is continuously annotated in frustration and satisfaction. This corpus has been set up to develop new systems able to model the continuous aspect of semantic and paralinguistic information at the conversation level. The present work focuses on the paralinguistic level, more precisely on the expression of emotions. In the call center industry, the conversation usually aims at solving the caller's request. As far as we know, most emotional databases contain static annotations in discrete categories or in dimensions such as activation or valence. We hypothesize that these dimensions are not task-related enough. Moreover, static annotations do not enable to explore the temporal evolution of emotional states. To solve this issue, we propose a corpus with a rich annotation scheme enabling a real-time investigation of the axis frustration / satisfaction. AlloSat regroups 303 conversations with a total of approximately 37 hours of audio, all recorded in real-life environments collected by Allo-Media (an intelligent call tracking company). First regression experiments, with audio features, show that the evolution of frustration / satisfaction axis can be retrieved automatically at the conversation level.

In 5G, low latency and high reliability have become the most significant issues in addition to ultra-high throughput and support of massive number of devices. To achieve the goal of low latency, the concept of tactile Internet has been introduced. To reduce the latency, the various techniques, which can be applied to from the physical layer to the application layer, are needed depending on the specific application and network architecture. In this special issue, the various topics handling such issues are included.

In blockchain, transactions between parties are regrouped into blocks, in order to be added to the blockchain’s distributed ledger. Miners are nodes of the network that generate new blocks according to the consensus protocol. The miner that adds a valid block to the distributed ledger is rewarded. However, to find a valid block, the miner needs to solve a computationally difficult problem, which makes it difficult to a single miner to gain rewards. Therefore, miners join mining pools, where the powers’ of miners are federated to ensure stable revenues. In public blockchains, access to mining pools is not restricted, which makes mining pools vulnerable to considerable threats such as: block withholding (BWH) attacks and distributed denial of service (DDoS) attacks. In the present work, we propose a new reputation based blockchain named PoolCoin based on a distributed trust model for a mining pools. The trust model used by PoolCoin is inspired from the job market signaling model. The proposed PoolChain blockchain allows pool managers the selection of trusted miners in their mining pools, while miners are able to evaluate them. Furthermore, to detect malicious miners that claim bigger computing capacity, we also provided a machine learning module to estimate the real miners’ capacities. The efficiency of the proposed trust model is studied and the obtained simulation results are presented and discussed. Thus, the model parameters’ are optimized in order to detect and exclude misbehaving miners, while honest miners are maintained in the mining pool.

In this paper, we present nmtpy, a flexible Python toolkit based on Theano for training Neural Machine Translation and other neural sequence-to-sequence architectures. nmtpy decouples the specification of a network from the training and inference utilities to simplify the addition of a new architecture and reduce the amount of boilerplate code to be written. nmtpy has been used for LIUM's top-ranked submissions to WMT Multimodal Machine Translation and News Translation tasks in 2016 and 2017., 10 pages, 3 figures

Lever les yeux est un prototype imagine au musee Dauphinois comme une experience entre contemplation et mediation. © Musee Dauphinois/Denis Vincon Visiter une exposition dans un espace de realite virtuelle ; arpenter, a l’aide de sa souris, une salle d’exposition reconstituee en profondeur sur son ecran d’ordinateur ; explorer une salle par robot interpose ou encore faire l’experience d’une œuvre grâce a un site Internet complementaire a l’exposition physique sont autant d’experiences de dec...

International audience; Apart from the the increasing demand of smartphones in human-to-human (H2H) communications, the introduction of machine-to-machine (M2M) devices poses significant challenges to wireless cellular networks. In order to offer the ability to connect billion of devices to propel the society into a new era of connectivity in our homes, officies and smart cities, we design novel radio resource sharing algorithms in a H2H/M2M coexistence case to accommodate M2M communications while not severely degrading existing H2H services. We propose group-based M2M communications that share the same spectrum with H2H communications through device-to-device (D2D) communication, as one of the technology components of 5G architecture. First, we formulate radio resource sharing problem as a sum-rate maximization, problem for which the optimal solution is non-deterministic polynomial-time hard (NP-hard). To overcome the computational complexity of the optimal solution, we model the resource sharing problem as a bipartite graph, then propose a novel interference-aware graph-based resource sharing scheme using a fixed M2M transmit power. To further enhance the protection of H2H services, we introduce an adaptive power control mechanism into the interference-aware graph-based resource sharing scheme. M2M transmit power is efficiently adjusted using one among the two following alternative controllers, namely, either the proportional integral derivative (PID) or the fuzzy logic. The latter is proposed within the aim to assure the desired quality-of-service (QoS) of H2H users and increase the efficiency of M2M spectrum usage. In both cases (fixed and adaptive), a centralized and a semi-distributed instantiations are given. Simulation results show that adaptive M2M radio resource sharing scheme using fuzzy logic is the one that achieves the best compromise. In fact, it guarantees H2H performance in terms of throughput and fairness while maximizing the efficiency of M2M spectrum usage. Simulation results also show that in spite of its quite good performance, semi-distributed M2M resource sharing instantiation achieves them with a decline of up to 10% in terms of H2H throughput compared to the centralized instantiation. This is achieved through a markedly lower communication overhead.

Nous présentons un nouveau corpus, nommé AlloSat, composé de conversations en français extraites de centre d’appels, annotées de façon continue en frustration et satisfaction. Dans le contexte des centres d’appels, une conversation vise généralement à résoudre la demande de l’appelant. Ce corpus a été mis en place afin de développer de nouveaux systèmes capables de modéliser l’aspect continu de l’information sémantique et para-linguistique au niveau conversationnel. Nous nous concentrons sur le niveau para-linguistique, plus précisément sur l’expression des émotions. À notre connaissance, la plupart des corpus émotionnels contiennent des annotations en catégories discrètes ou dans des dimensions continues telles que l’activation ou la valence. Nous supposons que ces dimensions ne sont pas suffisamment liées à notre contexte. Pour résoudre ce problème, nous proposons un corpus permettant une connaissance en temps réel de l’axe frustration/satisfaction. AlloSat regroupe 303 conversations pour un total d’environ 37 heures d’audio, toutes enregistrées dans des environnements réels, collectées par Allo-Media (une société spécialisée dans l’analyse automatique d’appels). Les premières expériences de classification montrent que l’évolution de l’axe frustration/satisfaction peut être prédite automatiquement par conversation.

La recherche d’acteurs vocaux pour les productions audiovisuelles est réalisée par des directeurs artistiques (DA). Les DA sont constamment à la recherche de nouveaux talents vocaux, mais ne peuvent effectuer des auditions à grande échelle. Les outils automatiques capables de suggérer des voix présentent alors un grand intérêt pour l’industrie audiovisuelle. Dans les travaux précédents, nous avons montré l’existence d’informations acoustiques permettant de reproduire des choix du DA. Dans cet article, nous proposons une approche à base de réseaux de neurones pour construire une représentation adaptée aux personnages/rôles visés, appelée p-vecteur. Nous proposons ensuite de tirer parti de données externes pour la représentation de voix, proches de celles d’origine, au moyen de méthodes de distillation de la connaissance. Les expériences menées sur des extraits de voix de jeux vidéo montrent une amélioration significative de l’approche p-vecteur, avec distillation de la connaissance, par rapport à une représentation x-vecteur, état-de-l’art en reconnaissance du locuteur.

National audience; La perte d’intelligibilité représente une plainte importante des patients traités pour un cancer de la cavité buccale ou de l’oropharynx. L’évaluation de l’intelligibilité est essentielle dans le parcours de soin, mais les tests existants ne sont pas satisfaisants. Basés sur la perception de listes de mots par des auditeurs entraînés à restaurer des séquences sonores dégradées, ils conduisent souvent à une sousévaluation des déficits. Nous avons proposé une nouvelle tâche d’évaluation de l’intelligibilité, la tâche de décodage acoustico phonétique (DAP), basée sur l’utilisation de pseudo-mots (Astésano et al., 2018; Ghio et al., 2018; Ghio et al., soumis, Lalain et al., sous presse). Dans cette étude, nous évaluons la capacité de la tâche DAP à neutraliser les effets de restauration lexicale et d’expertise auditive clinique. Les résultats montrent que contrairement à une évaluation de l’intelligibilité basée sur des mots, une évaluation basée sur des pseudo-mots permet d’obtenir des scores de Déviation Phonologique Perçue (DPP) stables au cours du temps quel que soit le degré d’expertise des auditeurs, naïfs ou cliniciens.

International audience; Extraction of semantic information from real-life speech, such as emotions, is a challenging task that has grown in popularity over the last few years. Recently, emotion processing in speech moved from discrete emotional categories to continuous affective dimensions. This trend helps in the design of systems that predict the dynamic evolution of affect in speech. However, no standard annotation guidelines exist for these dimensions thus making cross-corpus studies hard to achieve. Deep neural networks are nowadays predominant in the task of emotion recognition. Almost all systems use recurrent architectures, but convolutional networks were recently reassessed as they are faster to train and have less parameters than recurrent ones. This paper aims at investigating pros and cons of the aforementioned architectures using cross-corpus experiments to highlight the issue of corpus variability. We also explore the best suitable acoustic representation for continuous emotion, together with loss functions. We concluded that recurrent networks are robust to corpus variability and we confirm the power of cepstral features for continuous Speech Emotion Recognition(SER), especially for satisfaction prediction. A final post-treatment applied on prediction brings very nice result (ccc = 0.719) on AlloSat and achieves new state of the art.

La reconnaissance des entités nommées (REN) à partir de la parole est traditionnellement effectuée par l’intermédiaire d’une chaîne de composants, exploitant un système de reconnaissance de la parole (RAP), puis un système de REN appliqué sur les transcriptions automatiques. Les dernières données disponibles pour la REN structurées à partir de la parole en français proviennent de la campagne d’évaluation ETAPE en 2012. Depuis la publication des résultats, des améliorations majeures ont été réalisées pour les systèmes de REN et de RAP. Notamment avec le développement des systèmes neuronaux. De plus, certains travaux montrent l’intérêt des approches de bout en bout pour la tâche de REN dans la parole. Nous proposons une étude des améliorations en RAP et REN dans le cadre d’une chaîne de composants, ainsi qu’une nouvelle approche en trois étapes. Nous explorons aussi les capacités d’une approche bout en bout pour la REN structurées. Enfin, nous comparons ces deux types d’approches à l’état de l’art de la campagne ETAPE. Nos résultats montrent l’intérêt de l’approche bout en bout, qui reste toutefois en deçà d’une chaîne de composants entièrement mise à jour.