Search

International audience; This paper investigates the challenges posed by the application of reinforcement learning to large-scale strategy games. In this context, we present steps and techniques which synthesize new ideas with state-of-the-art techniques from several areas of machine learning in a novel integrated learning approach for this kind of games. The performance of the approach is demonstrated on the task of learning valuable game strategies for a commercial wargame.

International audience; Contract scheduling is a general technique that allows to design a system with interruptible capabilities, given an algorithm that is not necessarily interruptible. Previous work on this topic has largely assumed that the interruption is a worst-case deadline that is unknown to the scheduler. In this work, we study the setting in which there is a potentially erroneous prediction concerning the interruption. Specifically, we consider the setting in which the prediction describes the time that the interruption occurs, as well as the setting in which the prediction is obtained as a response to a single or multiple binary queries. For both settings, we investigate tradeoffs between the robustness (i.e., the worst-case performance assuming adversarial prediction) and the consistency (i.e, the performance assuming that the prediction is error-free), both from the side of positive and negative results.

International audience; This roundtable explores the role the EU could play at the Global level considering the sustainability, governance and regulation of the digital transformation. As digitalization harbours opportunities and challenges for democracy and for the governance of technological trajectories, it will discuss whether and how normative diffusion could take place, with European responses being taken up in other countries, especially from the Global South, and to what extent and under which conditions such responses could serve as models of reference. To this end, the roundtable will feature a discussion among colleagues from both EU and Emerging Powers, participating to the European project PRODIGEES (Promoting Research on Digitalisation in Emerging Powers and Europe towards Sustainable Development), aiming at fostering international collaboration and sharing of knowledge on (global) governance and the conditions to shape digitalisation towards the Sustainable Development Goals in the 2030 Agenda.The discussion will address various forms of regulation and governance of digital data and economic processes by democratic institutions, investigating issues such as: transparency and accountability rules such as those laid down in ‘open government partnerships’; citizens participation and engagement in political processes, as permitted by digital forms of communication; big data, raising concerns about social polarization, populism, and the fragmentation of the public sphere; platforms and data economies, considering disruptions of old business models and novel, asymmetric power relationships; and regulatory responses provided in Europe aiming at protecting fundamental rights, enabling a digital single market, and counteracting digital power asymmetries.

International audience; This paper is motivated by the repetitive and periodic transmission of messages in Wireless Sensor Networks (WSNs) given by the ZigBee standard. In order to save energy, communication tasks are grouped in time. The WSN applications, such as control loops used in production lines, impose deadlines on the message delivery time. By introducing a grouping constraint, this paper extends the polynomial cyclic scheduling problem of building a periodic schedule with uniform precedence constraints and no resource constraints. We consider that each task belongs to one group, and each group is to be scheduled as a super-task in every period. We show two examples issued from different applications of such grouping constraints, using uniform constraints to model the deadlines expressed in the number of periods. We tackle the feasibility problem (the existence of a periodic schedule), which has shown to be independent of the processing times. We propose a graph formulation of the problem using the retiming technique. In the ZigBee case, we prove that the particular tree structure of the groups and of the uniform precedences based upon the communication tree, lead to a polynomial algorithm to solve the problem. But the general problem is proven to be NP-complete even if no additional resource constraints are considered, and unit processing times are assumed. This extension of the cyclic scheduling leads to a new range of grouping problems with applications, not only in communication networks but also in production and logistics scheduling.

International audience; Convolutional neural networks are nowadays witnessing a major success in different pattern recognition problems. These learning models were basically designed to handle vectorial data such as images but their extension to non-vectorial and semi-structured data (namely graphs with variable sizes, topology, etc.) remains a major challenge, though a few interesting solutions are currently emerging.In this paper, we introduce MLGCN; a novel spectral Multi-Laplacian Graph Convolutional Network. The main contribution of this method resides in a new design principle that learns graph-laplacians as convex combinations of other elementary laplacians – each one dedicated to a particular topology of the input graphs. We also introduce a novel pooling operator, on graphs, that proceeds in two steps: context-dependent node expansion is achieved, followed by a global average pooling; the strength of this two-step process resides in its ability to preserve the discrimination power of nodes while achieving permutation invariance. Experiments conducted on SBU and UCF-101 datasets, show the validity of our method for the challenging task of action recognition.

International audience; To increase software performance, it is now common to use hardware accelerators. Currently, GPUs are the most widespread accelerators that can handle general computations. This requires to use GPGPU frameworks such as Cuda or OpenCL. Both are very low-level and make the benefit of GPGPU programming difficult to achieve. In particular, they require to write programs as a combination of two subprograms, and, to manually manage devices and memory transfers. This increases the complexity of the overall software design. The idea we develop in this paper is to guarantee expressiveness and safety for CPU and GPU computations and memory managements with high-level data-structures and static type-checking. In this paper, we present how statically typed languages, compilers and libraries help harness high level GPGPU programming. In particular, we show how we added high-level user-defined data structures to a GPGPU programming framework based on a statically typed programming language: OCaml. Thus, we describe the introduction of records and tagged unions shared between the host program and GPGPU kernels described via a domain specific language as well as a simple pattern matching control structure to manage them. Examples, practical tests and comparisons with state of the art tools, show that our solutions improve code design, productivity, and safety while providing a high level of performance.

International audience; This paper introduces a new encoder-decoder architecture that is trained to reconstruct images by disentangling the salient information of the image and the values of attributes directly in the latent space. As a result, after training, our model can generate different realistic versions of an input image by varying the attribute values. By using continuous attribute values, we can choose how much a specific attribute is perceivable in the generated image. This property could allow for applications where users can modify an image using sliding knobs, like faders on a mixing console, to change the facial expression of a portrait, or to update the color of some objects. Compared to the state-of-the-art which mostly relies on training adversarial networks in pixel space by altering attribute values at train time, our approach results in much simpler training schemes and nicely scales to multiple attributes. We present evidence that our model can significantly change the perceived value of the attributes while preserving the naturalness of images.

International audience; A Markov decision process (MDP) is a general model for solving planning problems under uncertainty. It has been extended to multiobjective MDP to address multicriteria or multiagent problems in which the value of a decision must be evaluated according to several viewpoints, sometimes conflicting. Although most of the studies concentrate on the determination of the set of Pareto-optimal policies, we focus here on a more specialized problem that concerns the direct determination of policies achieving well-balanced tradeoffs. To this end, we introduce a reference point method based on the optimization of a weighted ordered weighted average (WOWA) of individual disachievements. We show that the resulting notion of optimal policy does not satisfy the Bellman principle and depends on the initial state. To overcome these difficulties, we propose a solution method based on a linear programming (LP) reformulation of the problem. Finally, we illustrate the feasibility of the proposed method on two types of planning problems under uncertainty arising in navigation of an autonomous agent and in inventory management.

National audience; Collaborative online games, such as Massively Multiplayer Online Role-playing Games (MMORPG), can lead players to join groups, guilds or lead battles to achieve a shared goal by providing a world in which each one can have shared, bonding experiences. Our problem is to be able to formally model a multi-role scenario. The first question is “what is a multi-role scenario?” and the next one is “how to model its specific constraints?”. For that purpose, we based this initiating work on a real ad-hoc scenario that has been enacted in a serious game last year. Indeed, we tried to model it with several available tools and propose the results of this comparison in this paper. The objective of this paper is to find and/or identify relevant functionalities used for multi-role in multiplayer serious game scenario modelling; Topaze, TINA, and LEGADEE. From our case scenario modelling demonstration, we found that each tool had pros and cons.

International audience; We investigated the calibration of learners’ metacognitive judgments, content page visitation, and study time allocation in the context of learning about the human circulatory system with MetaTutor, a multi-agent hypermedia learning environment. Our participants were 100 college students. Both pedagogical agent (PA) and an open learner model (OLM) were used to provide metacognitive judgment feedback to participants. The results indicated that immediate feedback led to significantly lower over- and underconfidence in participants’ metacognitive judgments and more frequent and longer duration visitations of pages relevant to their current learning sub-goal. The OLM acted as an immediate feedback mechanism for participants who didn’t receive PA’s feedback. All participants achieved significantly higher learning out comes pre-to-posttest, with no difference between with-feedback and no-feedback conditions.

National audience; Dans cette communication, nous présentons une procédure originale et générale de calcul de borne inférieure pour des problèmes d'optimisation robuste avec coûts intervalles où l'on cherche à minimiser le regret maximal.

In a standard Markov decision process (MDP), rewards are assumed to be precisely known and of quantitative nature. This can be a too strong hypothesis in some situations. When rewards can really be modeled numerically, specifying the reward function is often difficult as it is a cognitively-demanding and/or time-consuming task. Besides, rewards can sometimes be of qualitative nature as when they represent qualitative risk levels for instance. In those cases, it is problematic to use directly standard MDPs and we propose instead to resort to MDPs with ordinal rewards. Only a total order over rewards is assumed to be known. In this setting, we explain how an alternative way to define expressive and interpretable preferences using reference points can be exploited.

This paper addresses handover decision instability which impacts negatively on both user perception and network performances. To this aim, a new technique called The HandOver Decision STAbility Technique (HODSTAT) is proposed for horizontal handover in Wireless Local Area Networks (WLAN) based on IEEE 802.11standard. HODSTAT is based on a hysteresis margin analysis that, combined with a utilitybased function, evaluates the need for the handover and determines if the handover is needed or avoided. Indeed, if a Mobile Terminal (MT) only transiently hands over to a better network, the gain from using this new network may be diminished by the handover overhead and short usage duration. The approach that we adopt throughout this article aims at reducing the minimum handover occurrence that leads to the interruption of network connectivity (this is due to the nature of handover in WLAN which is a break before make which causes additional delay and packet loss). To this end, MT rather performs a handover only if the connectivity of the current network is threatened or if the performance of a neighboring network is really better comparing the current one with a hysteresis margin. This hysteresis should make a tradeoff between handover occurrence and the necessity to change the current network of attachment. Our extensive simulation results show that our proposed algorithm outperforms other decision stability approaches for handover decision algorithm., 13 Pages, IJWMN

International audience; Motivated by the deployment of post-disaster MANEMO (MANET for NEMO) composed of mobile routers and stations, we evaluate two candidate routing protocols through network simulation, theoretical performance analysis, and field experiments. The first protocol is the widely adopted Optimized Link State Routing protocol (OLSR) and the second is the combination of the Tree Discovery Protocol (TDP) with Network In Node Advertisement (NINA). To the best of our knowledge, this is the first time that these two protocols are compared in both theoretical and practical terms. We focus on the control overhead generated when mobile routers perform a handover. Our results confirm the correctness and operational robustness of both protocols. More interestingly, although in the general case OLSR leads to better results, TDP/NINA outperforms OLSR both in the case of sparse networks and in highly mobile networks, which correspond to the operation point of a large set of post-disaster scenarios.

International audience; The methods of automatic classification resulting from the artificial intelligence are generally the consequences of a formalism based on an artificial reasoning quasi similar to that of the human expert. All the approaches of automatic classification developed so far, whether in an exact or approximate context, are dissociated from each other by the membership concept of an object to a class. In this paper, we present a new approach hybrid of unsupervised automatic classification under the C-Means (means of C classes) family. This new approach, based on the fusion of fuzzy and possibility theory and initialized by a membership matrix, allows on the one hand to solve simultaneously the problem of overlapping and coincidence, to reduce the noise effect and on the other hand to accelerate the classification process. The model validation is carried out by the FCM (Fuzzy C-Means), the PCM (Possibilistic C-Means)and the FPCM (Fuzzy-Possibilistic C-Means) for two cases of initialization by using Iris, Textured image and Tight human data basis.

Differences in simulation results may be observed from one architecture to another or even inside the same architecture. Such reproducibility failures are often due to different rounding errors generated by different orders in the sequence of arithmetic operations. Reproducibility problems are particularly noticeable on new computing architectures such as multicore processors or GPUs (Graphics Processing Units). DSA (Discrete Stochastic Arithmetic) enables one to estimate rounding error propagation in simulation programs. In this paper, it is shown that DSA can be used to estimate which digits in simulation results may be different from one environment to another because of rounding errors. A particular implementation of DSA, which enables numerical validation in hybrid CPU-GPU environments, is described. The estimation of numerical reproducibility using DSA is illustrated by a wave propagation code which can be affected by reproducibility problems when executed on different architectures.

International audience; Like most data models encountered in the Big Data ecosystem, RDF stores are managing large data sets by partitioning triples across a cluster of machines. Nevertheless, the graphical nature of RDF data as well as its associated SPARQL query execution model makes the efficient data distribution more involved than in other data models, e.g., relational. In this paper, we propose a novel system that is characterized by a trade-off between complexity of data partitioning and efficiency of query answering in cases where a query workload is known. The prototype is implemented over the Apache Spark framework, ensuring high availability, fault tolerance and scalability. This short paper presents the main features of the system and highlights the omnipresence of parallel computation across data fragmentation and allocation, encoding and query processing tasks.

International audience; In this paper a new on-line algorithm is proposed (the Droplets algorithm) for dealing with concept drifts and to produce reliable predictions. The two main characteristics of this algorithm are that it is able to adapt to different types of drifts without making any assumptions regarding their type or when they occur, and can provide reliable predictions in a non-stationary environment without using a fixed confidence threshold. Experimental results on five datasets based on Random RBF and Rotating Hyperplane generators as well as a new semi-synthetic dataset based weather temperatures show that, by discarding difficult observations, the Droplets algorithm manages to obtain the best average accuracy against ten classifiers. The results also indicate that the algorithm manages to provide reliable prediction by accurately distinguishing which observations are easily classifiable.

Querying very large RDF data sets in an efficient manner requires a sophisticated distribution strategy. Several innovative solutions have recently been proposed for optimizing data distribution with predefined query workloads. This paper presents an in-depth analysis and experimental comparison of five representative and complementary distribution approaches. For achieving fair experimental results, we are using Apache Spark as a common parallel computing framework by rewriting the concerned algorithms using the Spark API. Spark provides guarantees in terms of fault tolerance, high availability and scalability which are essential in such systems. Our different implementations aim to highlight the fundamental implementation-independent characteristics of each approach in terms of data preparation, load balancing, data replication and to some extent to query answering cost and performance. The presented measures are obtained by testing each system on one synthetic and one real-world data set over query workloads with differing characteristics and different partitioning constraints., 16 pages, 3 figures

International audience

We consider the following problem: two nodes want to reliably communicate in a dynamic multihop network where some nodes have been compromised, and may have a totally arbitrary and unpredictable behavior. These nodes are called Byzantine. We consider the two cases where cryptography is available and not available. We prove the necessary and sufficient condition (that is, the weakest possible condition) to ensure reliable communication in this context. Our proof is constructive, as we provide Byzantine-resilient algorithms for reliable communication that are optimal with respect to our impossibility results. In a second part, we investigate the impact of our conditions in three case studies: participants interacting in a conference, robots moving on a grid and agents in the subway. Our simulations indicate a clear benefit of using our algorithms for reliable communication in those contexts.

International audience; We study theoretical and practical aspects of five of the most well-known self-stabilizing dining philosophers algorithms. We theoretically prove that three of them are incorrect. For practical evaluation, we simulate all algorithms and evaluate their fault-tolerance, latency and throughput of critical section access. We present a new combined algorithm that achieves the best throughput of the two remaining correct algorithms by determining the system load and switching between these basic algorithms. We prove the combined algorithm correct, simulate it and study its performance characteristics.

International audience; Information cross-checking is an essential step of the trust building process that grants it its dynamics: it assesses the credibility dimension, finding confirmations or invalidations that respectively increase or weaken the current trust level of a considered piece of information and whose order influences its final value. This paper proposes a model of credibility integration that realistically takes into account even dubious confirmations and invalidations, allowing to represent a wide range of dynamic credulity stances when faced with contradictory information streams. It is formalised in an extended multivalued logic framework and illustrated with several examples to highlight the variety of behaviours it captures.

International audience; Since the World Summit on the Information Society’s Working Group on Internet Governance (2004-2005), there has been a growing emphasis on the use of the term multistakeholder in Internet governance (IG). The annual Internet Governance Forum established in 2006, the “NetMundial” jointly convened by ICANN and Brazil in May, 2014 and the recent and more controversial “NetMundial Initiative” launched as a follow-up with the World Economic Forum, highlight the need for multistakeholder discussions and consensus finding. They almost always delineate three main categories of stakeholders: governments, private sector, and civil society.However, we have shown in previous work that International Organizations (IOs) also are crafting roles for themselves as IG actors. Underlining the need for their invaluable experience, capacity and mandate to co-elaborate binding and non binding standards, they have been trying to establish themselves as key players in multistakeholder settings. With a focus on two regional organizations (OECD and Council of Europe) and a global UN agency (UNESCO), explored through a set of interviews of their key leaders and personnel and through the analysis of their main produced outcomes, this paper addresses how IOs have been successful in navigating the murky waters of IG and multistakeholderism. Using both qualitative and quantitative analysis of interview and archival data, we examine the emergent roles of IOs and the paths they forge in multistakeholder arenas. Such pathways include coalescing with other stakeholders around some issues of common interest, crafting research agendas, and allowing their Secretariats to smoothly circumvent some of their Member States positions and, as a result, producing positions and non-binding international instruments related to the most contentious IG issues. We explore in-depth how these outcomes have managed to overcome sometimes very conflicting positions among different stakeholders, analyzing critically the processes and means that IOs use to straddle hegemonic and resisting positions from other stakeholders, in particular by shaping dialogues and influencing best practices.We use a conceptual framework from political science, communication theory, and organizational sociology. Enriched with empirical results, our research provides an inside and heretofore unexplored view of the workings of IOs in the IG field, illuminating how they enhance their roles (and concomitant power) in a complex, multilayered, highly discursive and often uncertain internet governance ecosystem.

International audience; The past decade of Internet governance issues and opportunities has been tumultuous andtransformation-filled. Much work has been done on nation-states, new institutions such as ICANN or theInternet Governance Forum or even the private sector. Less work focuses on the roles of internationalorganizations, including long-standing regional international organizations such as the Council of Europe(CoE) and their interactions with other key actors including the European Union. This paper takes a relativelylong term view of international organizations and the European Union in the complex and uncertain internetgovernance ecosystem, beginning with the World Summit on the Information Society (WSIS) and its craftingof an institutional innovation, the Internet Governance Forum, convened for the first time in 2006 in Athens,Greece. Using case study data from our research project on international organizations involved in internetgovernance, it reports and analyzes findings from the Council of Europe (COE). Particular attention is paid tothe COE role in the multistakeholder approach to Internet governance, highlighted in WSIS outcomedocuments and especially to the trajectory of such roles and any accompanying tensions and possibilities.The paper tracks a non-linear trajectory from a regional international organization as a region-widecoordinating mechanism for nation-state members to a global stakeholder itself. In so doing, the paperprovides data regarding initial outcomes of this trajectory with special reference to power and possibilities ofregional international organizations in the decade ahead.

International audience; The aim of this paper is to propose a new approach interweaving preference elicitation and search to solve multiobjective optimization problems. We present an interactive search procedure directed by an aggregation function, possibly non-linear (e.g. an additive disutility function, a Choquet integral), defining the overall cost of solutions. This function is parameterized by weights that are initially unknown. Hence, we insert comparison queries in the search process to obtain useful preference information that will progressively reduce the uncertainty attached to weights. The process terminates by recommending a near-optimal solution ensuring that the gap to optimality is below the desired threshold. Our approach is tested on multiobjective state space search problems and appears to be quite efficient both in terms of number of queries and solution times.

International audience; MOOCs constitute an important evolution of the environments of e-learning. Despite the success of this phenomenon, the implementations of MOOC content show that many obstacles remain. Moreover, the educational resources are rather simple (course, exercise, blog...) and are stored in a dedicated space. The provision of external pedagogical resources (Serious Game, Remote Laboratory...) becomes crucial. These complex resources are essential to a coherent pedagogical path and to ensure the practical application and the evaluation of the skills required by the learner. In this work, we propose an approach allowing the use of pedagogical resources deployed in MOOC platforms. The production of resources requires consequent data-processing developments and the investment of time, money and human resources. It is fundamental to exploit them more than once and thus to be able to reuse and share them. In this context, the structuring of educational information via standards becomes crucial.LOM is a standard that allows indexing pedagogical resources. This standard is considered complex. Moreover, LOM specifies only four levels of aggregation, which are rather general and vague. On the evaluation level, LOM did not specify information for the validation of the learner's skills.The SCORM standard presents the aggregation of resources and not a sequence of activities, or any navigation from one to another and any communication with the MOOC. Moreover, it does not specify how a MOOC should use the various organization of the content. Concerning SCORM, it is not possible to describe the assessment action, which will be realized with the aggregate contents. It results in a lack of tracking of the learner's activity, that is though necessary for the evaluation of its competencies.In order to illustrate our remarks, we present here a complex resource “remote laboratory”. A remote laboratory proposes a practical activity that is necessary to the learner's formation. It consists in a shareable resource among learners and it is composed of several resources deployed on real physical hardware reachable via remote platforms. This resource is time-constrained: a remote laboratory lasts a fixed duration with an accessibility constraint due to the shareable property. In the case of complex and evolving resources over the time, those standards do not propose any time constraints on the resource and do not present some mutual exclusion concerning a shareable critical resource. We propose an integration model of a complex pedagogical resource in a MOOC platform. This model deals with static and dynamic aspects. The static elements are those that characterize the structure and the contextualization of educational resource. The dynamic components concern the services that the resource needs for its use. In this paper, we proposed an approach to integrate complex resources in MOOC (ClarolineConnect). The current resources present very often an insufficiency in practical application and then a lack of evaluation of the relevant competencies acquired by the learner. In this context, we focused on the problem and we built a bundle allowing to deploy a remote laboratory resource. This resource is obviously complex: it has a reservation module of the different resources used and relies also on real hardware. These resources, digital and not, are deployed in distant platforms and invocated from ClarolineConnect via web services.

International audience; In this paper we adopt Skew Symmetric Bilinear (SSB) utility functions to compare policies in Markov Decision Processes (MDPs). By considering pairs of alternatives, SSB utility theory generalizes von Neumann and Morgenstern’s expected utility (EU) theory to encompass rational decision behaviors that EU cannot accommodate. We provide a game-theoretic analysis of the problem of identifying an SSB-optimal policy in finite horizon MDPs and propose an algorithm based on a double oracle approach for computing an optimal (possibly randomized) policy. Finally, we present and discuss experimental results where SSB-optimal policies are computed for a popular TV contest according to several instantiations of SSB utility functions.

National audience; In this paper we discuss the methodology and results of aligning three different emotional measurement methods (automatic facial expression recognition, self-report, electrodermal activation) and their agreement regarding learners’ emotions. Data was collected from 67 undergraduate students from a North American university who interacted with {MetaTutor}, an intelligent, multi-agent, hypermedia environment for learning about the human circulatory system, for a 1 hour learning session (Azevedo et al., 2013, Harley, Bouchet, \& Azevedo, 2013). A webcam was used to capture videos of learners’ facial expressions, which were analyzed using automatic facial recognition software ({FaceReader} 5.0). Learners’ physiological arousal was measured using Affectiva’s Q-Sensor 2.0 electrodermal activation bracelet. Learners self-reported their experience of 19 different emotional states (including basic, learner-centered, and academic achievement emotions) using the Emotion-Value questionnaire (Harley et al., 2013). They did so on five different occasions during the learning session, which were used as markers to align data from {FaceReader} and Q-Sensor. We found a high agreement between the facial and self-report data (75.6%) when similar emotions were grouped together along theoretical dimensions and definitions (e.g., anger and frustration) (Harley, et al., 2013). However, our new results examining the agreement between the Q-Sensor and these two methods suggests that electrodermal ({EDA}/physiological) indices of emotions do not have a tightly coupled (Gross, Sheppes, \& Urry, 2011) relationship with them. Explanations for this finding are discussed.

National audience

National audience; En théorie de la Décision, dans le cadre de la planification, l'approche factorisée des Processus Décisionnels Markoviens (FMDP) a débouché sur des algorithmes efficaces utilisant les arbres de décision (SVI, SPI) ou les diagrammes de décision algébriques (SPUDD).Pouvoir apprendre de tels modèles factorisés d'environnements stochastiques offre un avantage certain lors de la recherche de politique optimale.Des approches telles que l'architecture SDYNA proposent d'intégrer l'apprentissage et la planification.Toutefois, l'état de l'art pour l'apprentissage incrémental, pour la planification théorique de la décision structurée ou pour l'apprentissage par renforcement requièrent que le problème soit caractérisé à l'aide de variables binaires et/ou utilise des structures de données améliorables en termes de compacité.Magnan & al. propose l'utilisation de diagrammes de décision multimodaux (MDDs) comme structures de données plus efficaces et décrit des algorithmes de planification dédiés à ces structures.Cet article s'intéresse à l'apprentissage de tels modèles factorisés.Son principal résultat est l'algorithme IMDDI d'apprentissage en-ligne de MDDs qui offre des améliorations significatives de compacité du modèle et de temps d'apprentissage. De plus, aucune connaissance a priori sur le nombre de modalités des variables aléatoires n'est requise -- propriété intéressante pour un algorithme d'apprentissage en-ligne qui le démarque également de l'état de l'art.Enfin, cet algorithme permet une toute nouvelle instanciation de l'architecture SDYNA exploitant les MDDs : l'instance SPIMDDI.

International audience; Given a graph G=(V,E)G=(V,E) with |V|=n|V|=n and |E|=m|E|=m, we consider the metric cone MET(G) and the metric polytope METP(G ) defined on RERE. These polyhedra are relaxations of several important problems in combinatorial optimization such as the max-cut problem and the multicommodity flow problem. They are known to have non-compact formulations via the cycle inequalities in the original space RERE and compact (i.e. polynomial size) extended formulations via the triangle inequalities defined on the complete graph KnKn. In this paper, we show that one can reduce the number of triangle inequalities to O(nm)O(nm) and still have extended formulations for MET(G) and METP(G ). This is particularly interesting for sparse graphs when m=O(n)m=O(n).

International audience; Graph analysis is an essential tool to understand natural and man-made networks, such as social networks, food webs, transportation infrastructures, etc. Although graph analysis has fomented the development of algorithms, visual tools, and distributed processing frameworks, there is still little support for analysis at the query language level. Current graph query languages are mostly concerned with flexible matching of subgraphs, while graph processing frameworks are mostly concerned with fast parallel execution of instructions.Our goal is to provide analysis capabilities at the language level, allowing more interactive and explorative query-based analysis. In this paper, we present our ongoing efforts towards a relational algebra extension that offers an operator for graph-based data aggregation. The beta (β) operator is composed of four suboperators, which are used to control the path-based aggregations. The β-algebra allows seamless composition of queries that mix relational and graph-based aspects.Here we introduce our current algebra and provide examples of its use. We also show how we are using the analysis strategy in query scenarios. Since the algebra-based query scenario allows for execution plan rewritings, we also discuss our first efforts on equivalence rules for query optimization.

International audience

International audience; The SystemC AMS extensions support heterogeneous modeling and make use of several Models of Computation (MoCs) that operate on different time scales in the Discrete Event (DE), Discrete Time (DT), and Continuous Time (CT) domains. The simulation of such heterogeneous models may raise synchronization problems that are hard to diagnose and to fix, especially when considering multi-rate data flow parts. In this paper, we show how to formally analyze the execution of Timed Data Flow (TDF) models including their interaction with the DE domain by converting the synchronization mechanics into a Coloured Petri Net (CPN) equivalent. The developed symbolic execution algorithm for the CPN allows to detect all DE-TDF synchronization issues before simulation and to propose appropriate sample delay settings for the TDF converter ports to make the system schedulable. The presented technique is validated with a case study including a vibration sensor model and its digital front end.

International audience

International audience; Discussing results of our joint project that examines the complexinteractions among intergovernmental organizations and othertransnational institutions and nonstate actors in the global Internetgovernance ecosystem, this study highlights themes related to thechanging architecture and roles of international organizations from WSIS(World Summit on the Information Society) until NetMundial. Attention ispaid to old and new categories of organizations that emerged in thiscontext; and how they have been recognized as stakeholders in theprocess. These organizations form a network, set in an environmentalcontext, thus constituting the interorganizational infrastructure for internetgovernance today. Additionally, tracing knowledge flows and powerdifferentials over time among the different stakeholder organizations helpsto illustrate a major finding, the pro-active role of the internationalorganizations studied here in the messy, complex, and cross-nationalinternet governance ecosystem, shaped by and, at the same time, shapingthe technical infrastructure.

Belief revision games (BRGs) are concerned with the dynamics of the beliefs of a group of communicating agents. BRGs are "zero-player" games where at each step every agent revises her own beliefs by taking account for the beliefs of her acquaintances. Each agent is associated with a belief state defined on some finite propositional language. We provide a general definition for such games where each agent has her own revision policy, and show that the belief sequences of agents can always be finitely characterized. We then define a set of revision policies based on belief merging operators. We point out a set of appealing properties for BRGs and investigate the extent to which these properties are satisfied by the merging-based policies under consideration.

National audience

This chapter presents a platform for modeling, design, optimization, and co-simulation of mixed-signal systems using the SystemC-AMS standard. The platform is based on a bottom-up design and top-down simulation methodologies. In the bottom-up design methodology, an optimizer is inserted to perform a knowledge-aware optimization loop. During the process, a PEANO trajectory is applied for global exploration and the Nelder–Mead simplex optimization method is applied for local refinement. The authors introduce an interface between system-level models and their circuit-level realizations in the proposed platform. Moreover, a transient simulation scheme is proposed to simulate nonlinear dynamic behavior of complete mixed-signal systems. The platform is used to design and verify a low-power CMOS voltage regulator for an implantable telemetry system.

We consider the problem of gathering n autonomous robots that evolve in a 2-dimensional Euclidian space at a single location, not known beforehand. We suppose the robots operate in the semi-synchronous Look-Compute-Move model and are anonymous, oblivious, and disoriented. When robots are capable of strong multiplicity detection that is, sensing the number of robots at a given location and the initial configuration is not bivalent, the problem is known to be deterministically solvable for n>2 robots. When an arbitrary number f

National audience

International audience

International audience; Today, most students enrolling in a MOOC already have attended to another MOOC before. We study here a subcategory of these students, who register to the same MOOC several times, which we call Recurring Students (RS). Using data collected during three 5-week sessions of the GdP MOOC (N {\textgreater} 14,000 on average per session), we show there is a significant and increasing proportion (over 10%) of RS. While their re-enrollment seems influenced by changes in course content, they succeed in the same proportions or less than new students (NS). A more granular analysis, separating RS who had previously completed a track (Recurring Successful Students - RSS) from those who hadn’t (Recurring Unsuccessful Students - RUS), reveals that RUS complete the new session in the same proportion as NS, whereas RSS fail more. As a conclusion, we propose recommendations to address these students, which could be valuable for other multiple-session MOOCs.

International audience

This paper proposes incremental preference elicitation methods for multiobjective state space search. Our approach consists in integrating weight elicitation and search to determine, in a vector-valued state-space graph, a solution path that best fits the Decision Maker's preferences. We first assume that the objective weights are imprecisely known and propose a state space search procedure to determine the set of possibly optimal solutions. Then, we introduce incremental elicitation strategies during the search that use queries to progressively reduce the set of admissible weights until a nearly-optimal path can be identified. The validity of our algorithms is established and numerical tests are provided to test their efficiency both in terms of number of queries and solution times.

International audience

International audience; Pedagogical agents (PAs) have the ability to scaffold and regulate students' learning about complex topics while using intelligent tutoring systems (ITSs). Research on ITSs predominantly focuses on the impact that these systems have on overall learning, while the specific components of human-ITS interaction, such as student-PA dialogue within the system, are given little attention. One hundred undergraduate students interacted with MetaTutor, a multiagent hypermedia ITS, to learn about the human circulatory system. Data from these interactions were drawn from questionnaires and log-files to determine the extent to which a specific agent from MetaTutor, Sam the Strategizer, impacted students' overall emotions while using the system. Results indicated that Sam negatively impacted students' experiences of enjoyment, in relation to the other agents of MetaTutor, and the frequency of Sam's interactions with students significantly predicted their reports of boredom while using the system. Implications for the design of affect-sensitive multiagent ITSs are discussed.

National audience; L'avènement des systèmes à ressources restreintes dans notre quotidien exige des garanties sur leur fonctionnement. Parmi celles-ci, la consommation mémoire est un des éléments essentiels. Il est intéressant d'utiliser des mécanismes d'allocation et de récupération dynamique de mémoire pour une utilisation optimale de celle-ci. Mais cette dynamicité rend difficile la prévision de la quantité maximale de mémoire nécessaire au bon fonctionnement d'un programme. Pour permettre cette prévision, on propose le développement conjoint d'un langage à la ML muni d'un mécanisme de régions et d'une analyse fournissant une borne supérieure de la quantité de mémoire consommée. Cette analyse exploite les annotations de régions pour tenir compte de la mémoire libérée à travers celles-ci et ainsi affiner la borne supérieure de son occupation Comme le style de programmation influence les dépendances entre régions, il impacte aussi le résultat de l'analyse. Au final, on parvient à fournir une borne supérieure pour des programmes fonctionnels et impératifs manipulant des structures linéaires et arborescentes.

International audience; This paper deals with the question of the transfer of the diagnostic assessment Pépite at different grade levels. First, we present the theoretical foundations and the modalities of the diagnostic assessment Pépite. Then, we characterize the model of the diagnostic test and we study how this model is compatible according to different grade levels. We detail the response analysis and also we explain how to transfer it to different grade levels.