Search

TRA 2022, Transport Research Arena, Lisbon, PORTUGAL, 14-/11/2022 - 17/11/2022; In this paper, we address the railway safety assurance issues, impacting the conventional railway safety process due to the introduction of autonomy and AI technologies. Concretely, based on the system hierarchy and the railway safety lifecycle, we elaborate a methodology for the safety assurance of autonomous train and its operations. The proposed methodology consists of a set of safety processes and activities to be conducted, in parallel to the conventional ones, at three hierarchical system levels ; namely: (i) the overall system level (i.e., train); (ii) the AI-based component level (i.e., perception and decision-making components); (iii) and the AI software level (i.e., inference and decision learned algorithm).

International audience; Recently, the majority of successful matching approaches are based on convolutional neural networks, which focus on learning the invariant and discriminative features for individual image patches based on image content. However, the image patch matching task is essentially to predict the matching relationship of patch pairs, that is, matching (similar) or non-matching (dissimilar). Therefore, we consider that the feature relation (FR) learning is more important than individual feature learning for image patch matching problem. Motivated by this, we propose an element-wise FR learning network for image patch matching, which transforms the image patch matching task into an image relationship-based pattern classification problem and dramatically improves generalization performances on image matching. Meanwhile, the proposed element-wise learning methods encourage full interaction between feature information and can naturally learn FR. Moreover, we propose to aggregate FR from multilevels, which integrates the multiscale FR for more precise matching. Experimental results demonstrate that our proposal achieves superior performances on cross-spectral image patch matching and single spectral image patch matching, and good generalization on image patch retrieval.

International audience; The student's interaction with Virtual Learning Environments produces a large amount of data, known as learning traces, which is commonly used by the Learning Analytics (LA) domain to enhance the learning experience. We propose to define personas, that are representative of subsets of students sharing common digital behaviors. The embodiment of the output of LA systems in the form of personas makes it possible to study the representativeness of the dataset with precision and act accordingly, but also to enhance the explicability to pedagogical experts who must manipulate these tools. These personas are defined from learning traces, which are processed to identify homogeneous subsets of learners. The presented methodology also allows to identify some outliers, that exhibit atypical behaviors, and thus makes it possible to represent the whole students, without privileging some of them.

International audience; A newborn is preterm if birth occurred before a gestational age of 37 weeks. He has several immature functions, which implies a specific monitoring and, among others, the analysis of its sleep. Here we make a focus on Quiet Sleep (QS), whose increasing is primordial with age, and characterized by an absence of motion and a regular cardiorespiratory rhythm. A method to automatically detect QS is proposed, on the basis of a video analysis (detection of motion), supplemented by the estimation of ECG and respiration 'qualities'. This approach combines feature extraction and machine learning methods. It was validated on a set of 15 newborns and 25 eight-hours recordings manually annotated. Best results were obtained by combining non-motion intervals and ECG quality, but showing also an overestimation of QS (Se=88%, Sp=49%). However, regarding extracted features, we observed similar trends between manual and automated QS, with an increasing of average duration of QS intervals and percentage of time in QS with age, also approaching values of the full-term newborns. Finally, computation of QS on a larger set of 45 recordings confirmed the interest of the approach for maturation evaluation purposes. © 2021 Creative Commons.

International audience; Radio Frequency (RF) combined with the deep learning methods promised a solution to detect the presence of the drones. Indeed, the classical techniques (i.e. radar, vision and acoustics, etc.) suffer several drawbacks such as difficult to detect the small drones, false alarm of flying birds or balloons, the influence of the wind on the performance, etc. For an effective drones's detection, two main stages should be established: Feature extraction and feature classification. The proposed approach in this paper is based on a novel feature extraction method and an optimized deep neural network (DNN). At first, we present a novel method based on Welch to extract meaningful features from the RF signal of drones. Later on, three optimized Deep Neural Network (DNN) models are considered to classify the extracted features. The first DNN model can be used to detect the presence of the drones and contains two classes. The second DNN help us to detect and recognize the type of the drone with 4 classes: A class for each drone and the last one for the RF background activities. In the third model, 10 classes have been considered: the presence of the drone, its type, and its flight mode (i.e. Stationary, Hovering, flying with or without video recording). Our proposed approach can achieve an average accuracy higher than 94% and it significantly improves the accuracy, up to 30%, compared to existing methods.

International audience; Cetacean Distribution Modeling (CDM) is used to quantify mobile marine species distributions and densities. It is essential to better understand and protect whales and their relatives. Current CDM approaches often fail in capturing general species-environment relationships, which would be valid within a broader range of environmental conditions that characterize the surveyed regions. This paper aims at investigating the usefulness of deep learning based schemes, namely multi-task and transfer learning, in CDM. Co-training of a stochastic presence-background model on a classification task and a deterministic rule-based model on a regression task was performed. Whale presence-only records were used for the first task, and index outputs of a feeding habitat occurrence model for the second one. This new approach has been experimented through the study case of fin whales in the western Mediterranean Sea. To evaluate our approach, a new metric called True Positive rate per unit of Surface Habitat (TPSH) and an original multimodal fully-connected neural networks were developed. A Generalized Additive Model (GAM)—a standard CDM method—was also used as a reference for performance. Results show that our multi-task learning model improves both the feeding habitat model by 10.8% and data-driven models such as GAM by 16.5% on our TPSH metric in relative terms, revealing a higher accuracy of our approach in estimating whale presence. Such trends in results have been further supported by the use of two other independent datasets that forced models to generalize beyond their training dataset of species-environment relationships. Performance could be further improved by adopting more optimal thresholds as observed from Receiver Operating Characteristic curves, e.g. the multi-task learning model could reach absolute gains up to 10% in the median of the True Positive Rate while maintaining its habitat spatial spreading. Globally, our work confirmed our working hypothesis that expert information on whale behaviour represent a good knowledge base for model generalization. This result can be further improved by a concurrent learning of more local species-environment relationships from in-situ presence data.

International audience; Due to traffic conditions that are very context dependent, the computation of optimized or shortest paths is a very complex problem for both drivers and autonomous vehicles. In this paper, we introduce a learning mechanism that is able to efficiently evaluate path durations based on an abstraction of the available traffic information. We demonstrate that a cache data structure allows a permanent access to the results whereas a lazy politics taking new data into account is used to increase the viability of those results. Our measures highlight the performance of each mechanism, according to different learning strategies.

Modeling the dynamics of epidemics helps to propose control strategies based on pharmaceuticaland non-pharmaceutical interventions (contact limitation, lockdown, vaccination,etc). Hand-designing such strategies is not trivial because of the number of possibleinterventions and the difficulty to predict long-term effects. This task can be cast as an optimization problem where state-of-the-art machine learning methods such as deep reinforcement learning might bring significant value. However, the specificity of each domain|epidemic modeling or solving optimization problems|requires strong collaborationsbetween researchers from different fields of expertise. This is why we introduce EpidemiOptim, a Python toolbox that facilitates collaborations between researchers inepidemiology and optimization. EpidemiOptim turns epidemiological models and cost functions into optimization problems via a standard interface commonly used by optimization practitioners (OpenAI Gym). Reinforcement learning algorithms based on QLearning with deep neural networks (DQN) and evolutionary algorithms (NSGA-II) are already implemented. We illustrate the use of EpidemiOptim to find optimal policies fordynamical on-o lockdown control under the optimization of the death toll and economic recess using a Susceptible-Exposed-Infectious-Removed (SEIR) model for COVID-19. Using EpidemiOptim and its interactive visualization platform in Jupyter notebooks, epidemiologists, optimization practitioners and others (e.g. economists) can easily compare epidemiological models, costs functions and optimization algorithms to address important choicesto be made by health decision-makers. Trained models can be explored by experts and non-experts via a web interface. This article is part of the special track on AI and COVID-19.

International audience; In this paper a PD controller with intelligent compensation is used to solve the problem of tracking trajectories for a Delta Parallel Robot with three degrees of freedom. This controller uses an artificial B-Spline neural network as a feedforward compensation term. To evaluate the proposed controller performance some numerical simulations under two different scenarios have been carried out in order to know its effectiveness respect to a simple PD controller.

International audience; In this paper a PD controller with intelligent compensation is used to solve the problem of tracking trajectories for a Delta Parallel Robot with three degrees of freedom. This controller uses an artificial B-Spline neural network as a feedforward compensation term. To evaluate the proposed controller performance some numerical simulations under two different scenarios have been carried out in order to know its effectiveness respect to a simple PD controller.

International audience; The essential target of ‘Big Data’ technology is to provide new techniques and tools to assimilate and store large amount of generated data in a way to analyze and process it to get insights and predictions that can offer new opportunities towards the improvement of our life in different domains. In this context, ‘Big Data’ treats two essential issues: the real-time analysis issue introduced by the increasing velocity at which data is generated, and the long-term analysis issue introduced by the huge volume of stored data. To deal with these two issues, we propose in this paper a Cloud-based solution for big data analytic on Amazon Cloud operator. Our objective is to evaluate the performance of Big Data services offered regarding the volume/velocity of the processed data. The dataset we use contains information about”network connections” in approximately 5 million records with 41 features; the solution works as a network intrusion detector. It receives data records in real time from a raspberry pi node and predicts if the connection is bad (malicious intrusion or attack) or good (normal connection). The prediction model was made using a logistic regression network. We evaluate the cloud resources needed to train the machine learning model (batch processing), and to predict the new streaming data with the trained network in real time (real time processing). The solution worked very well with high accuracy and the results show that when working with Big Data in the cloud, we are mainly dealing with a cost/performance trade-off, the processing performance in term of response time for both long-term and real-time analysis can be always guaranteed once the cloud resources are well provisioned according to the needs.

International audience; Modern network intrusion detection systems rely on machine learning techniques to detect traffic anomalies and thus intruders. However, the ability to learn the network behaviour in real-time comes at a cost: malicious software can interfere with the learning process, and teach the intrusion detection system to accept dangerous traffic. This paper presents an intrusion detection system (IDS) that is able to detect common network attacks including but not limited to, denial-of-service, bot nets, intrusions, and network scans. With the help of the proposed example IDS, we show to what extent the training attack (and more sophisticated variants of it) has an impact on machine-learning based detection schemes, and how it can be detected.

Conference of European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2017 ; Conference Date: 18 September 2017 Through 22 September 2017; Conference Code:209269; International audience; We present WHODID: a turnkey intuitive web-based interface for fault detection, identification and diagnosis in production units. Fault detection and identification is an extremely useful feature and is becoming a necessity in modern production units. Moreover, the large deployment of sensors within the stations of a production line has enabled the close monitoring of products being manufactured. In this context, there is a high demand for computer intelligence able to detect and isolate faults inside production lines, and to additionally provide a diagnosis for maintenance on the identified faulty production device, with the purpose of preventing subsequent faults caused by the diagnosed faulty device behavior. We thus introduce a system which has fault detection, isolation, and identification features, for retrospective and on-the-fly monitoring and maintenance of complex dynamical production processes. It provides real-time answers to the questions: " is there a fault? " , " where did it happen? " , " for what reason? ". The method is based on a posteriori analysis of decision sequences in XGBoost tree models, using recurrent neural networks sequential models of tree paths. The particularity of the presented system is that it is robust to missing or faulty sensor measurements, it does not require any modeling of the underlying, possibly exogenous manufacturing process, and provides fault diagnosis along with confidence level in plain English formulations. The latter can be used as maintenance directions by a human operator in charge of production monitoring and control.

International audience; This book constitutes the refereed joint proceedings of the First International Workshop on Graphs in Biomedical Image Analysis, GRAIL 2017, the 6th International Workshop on Mathematical Foundations of Computational Anatomy, MFCA 2017, and the Third International Workshop on Imaging Genetics, MICGen 2017, held in conjunction with the 20th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2017, in Québec City, QC, Canada, in September 2017.

Conference of 9th IEEE International Memory Workshop, IMW 2017 ; Conference Date: 14 May 2017 Through 17 May 2017; Conference Code:128262; International audience; In the last decade machine learning algorithms have proven unprecedented performance to solve many real-world detection and classification tasks, for example in image or speech recognition. Despite these advances, there are still some deficits. First, these algorithms require significant memory access thus ruling out an implementation using standard platforms (e.g. GPUs, FPGAS) for embedded applications. Second, most machine leaning algorithms need to be trained with huge data sets (supervised learning). Resistive memories (RRAM) have demonstrated to be a promising candidate to overcome both these constrains. RRAM arrays can act as a dot product accelerator, which is one of the main building blocks in neuromorphic computing systems. This approach could provide improvements in power and speed with respect to the GPU-based networks. Moreover RRAM devices are promising candidates to emulate synaptic plasticity, the capability of synapses to enhance or diminish their connectivity between neurons, which is widely believed to be the basis for learning and memory in the brain. Neural systems exhibit various types and time periods of plasticity, e.g. synaptic modifications can last anywhere from seconds to days or months. In this work we proposed an architecture that implements both Short-And Long-Term Plasticity rules (STP and LTP) using RRAM arrays. We showed the benefits of utilizing both kinds of plasticity with two different applications, visual pattern extraction and decoding of neural signals. LTP allows the neural networks to learn patterns without training data set (unsupervised learning), and STP makes the learning process very robust against environmental noise.

International audience

Conference of 50th Annual IEEE International Carnahan Conference on Security Technology, ICCST 2016 ; Conference Date: 24 October 2016 Through 27 October 2016; Conference Code:125934; International audience; Customs administrations are responsible for the enforcement of fiscal integrity and security of movements of goods across land and sea borders. In order to verify whether the transported goods match the transport declaration, X-ray imaging of containers is used at many customs site worldwide. The main objective of the research and development project 'Automated Comparison of X-ray Images for Cargo Scanning (ACXIS)', which is funded by the European 7th Framework Program, is to improve the efficiency and effectiveness of the inspection procedures of cargo at customs using X-ray technology. The current inspection procedures are reviewed to identify risks, catalogue illegal cargo, and prioritize detection scenarios. Based on these results, we propose an integrated solution that provides automation, information exchange between customs administrations, and computer-based training modules for customs officers. Automated target recognition (ATR) functions analyze the X-ray image after a scan is made to detect certain types of goods such as cigarettes, weapons and drugs in the freight or container. Other helpful information can also be provided, such as the load homogeneity, total or partial weight, or the number of similar items. The ATR functions are provided as an option to the user. The X-ray image is transformed into a manufacturer-independent format through geometrical and spectral corrections and stored into a database along with the user feedback and other related data. This information can be exchanged with similar systems at other sites, thus facilitating information exchange between customs administrations. The database is seeded with over 30'000 examples of legitimate and illegal goods. These examples are used by the ATR functions through machine learning techniques, which are further strengthened by the information exchange. In order to improve X-ray image interpretation competency of human operators (customs officers), a computer-based training software is developed that simulates these new inspection procedures. A study is carried out to validate the effectiveness and efficiency of the computer-based training as well as the implemented procedures. Officers from the Dutch and Swiss Customs administrations partake in the study, covering both land and sea borders.

Conference of 33rd International Conference on Machine Learning, ICML 2016 ; Conference Date: 19 June 2016 Through 24 June 2016; Conference Code:124527; International audience; Let us consider a specific action or situation (called event) that takes place within a time series. The objective in early detection is to build a decision function that is able to go off as soon as possible from the onset of an occurrence of this event. This implies making a decision with an incomplete information. This paper proposes a novel framework that i) guarantees that a de-tection made with a partial observation will also occur at full observation of the time-series; ii) incorporates in a consistent manner the lack of knowledge about the minimal amount of information needed to make a decision. The proposed detector is based on mapping the temporal sequences to a landmarking space thanks to appropriately designed similarity functions. As a by-product, the framework benefits from a scalable training algorithm and a theoretical guarantee concerning its generalization ability. We also discuss an important improvement of our framework in which decision function can still be made reliable while being more expressive. Our experimental studies provide compelling results on toy data, presenting the trade-off that occurs when aiming at accuracy, earliness and reliability. Results on real physiological and video datasets show that our proposed approach is as accurate and early as state-of-the-art algorithm, while ensuring reliability and being far more efficient to learn.

International audience; This paper presents the “Multi-Role Project” method (MRP), a broadly applicable project-based learning method, and describes its implementation and evaluation in the context of a Science, Technology, Engineering, and Mathematics (STEM) course. The MRP method is designed around a meta-principle that considers the project learning activity as a role-playing game based on two projects: a learning project and an engineering project. The meta-principle is complemented by five principles that provide a framework to guide the working practices of student teams: distribution of responsibilities; regular interactions and solicitations within the team; anticipation and continuous improvement; positive interdependence and alternating individual/collective work; and open communication and content management. This paper presents the implementation of MRP in a course teaching software engineering, UML language, and project management. The results show that MRP helped the course's students to acquire important professional knowledge and skills, experience near-real-world professional realities, and develop their abilities to work both in teams and autonomously.; Cet article présente la méthode Multi-Role Project (MRP), une méthode d'apprentissage par projet largement applicable, et décrit son application et son évaluation de la contexte de cours en science, technologie, ingénierie et mathématiques. La méthode MRP est basée sur un méta-principe que considère l'activité d'apprentissage par projet comme un jeu de rôle basé sur deux projets : un projet d'apprentissage et un projet d'ingénierie. Le méta-principe est complété par cinq principes qui fournissent un cadre pour guider les pratiques de travail des équipes d'étudiants : la distribution des responsabilités, les sollicitations et interactions avec l'équipe, l'anticipation et l'amélioration continue, l'interdépendance positive et l'alternance du travail individuel et du travail collectif, et la gestion ouverte des contenus et de la communication? Cet article présente l'application de MRP dans un enseignement d'ingénierie logicielle, de modélisation UML et de gestion de projet. Les résultats montrent que MRP a aidé les étudiants à acquérir des compétences et connaissances professionnelles importantes, une expérience proche des réalités professionnelles, et à développer leurs capacités à travailler à la fois en équipe et en autonomie.

International audience; Deep brain stimulation of the medial globus pallidus is a surgical procedure for treating patients suffering from Parkinson's disease. Its therapeutic effect may be limited by the presence of pyramidal tract side effect (PTSE). PTSE is a contraction time-locked to the stimulation when the current spreading reaches the motor fibers of the pyramidal tract within the internal capsule. The lack of side-effect predictive model leads the neurologist to secure an optimal electrode placement by iterating clinical testing on an awake patient during the surgical procedure. The objective of the study was to propose a preoperative predictive model of PTSE. A machine learning based method called PyMAN (for Pyramidal tract side effect Model based on Artificial Neural network) that accounted for the current of the stimulation, the 3D electrode coordinates and the angle of the trajectory, was designed to predict the occurrence of PTSE. Ten patients implanted in the medial globus pallidus have been tested by a clinician to create a labeled dataset of the stimulation parameters that trigger PTSE. The kappa index value between the data predicted by PyMAN and the labeled data was.78. Further evaluation studies are desirable to confirm whether PyMAN could be a reliable tool for assisting the surgeon to prevent PTSE during the preoperative planning. © 2016 SPIE.

International audience; Organisational capability management is a recent approach; it enables global and coordinated management of resources and facilitates future synergies within extended organisations. This paper provides a state of the art of organisational capability approach and proposes a management framework. On the one hand, an overview of the theoretical principles and the practical solutions, especially methods for standardising and transferring firms' good practices, points out the limits of their sustainable functioning. On the other hand, the framework manages organisational capabilities by following the theoretical principles and overcoming identified barriers. This proposition is structured around three kinds of processes: primary (design and transfer), support (assessment and enhancement) and management (coordination and alignment).

Conference of 20th International Conference on Logic for Programming, Artificial Intelligence, and Reasoning, LPAR 2015 ; Conference Date: 24 November 2015 Through 28 November 2015; Conference Code:158539; International audience; Abstract interpretation is a powerful tool in program verification. Several commercial or industrial scale implementations of abstract interpretation have demonstrated that this approach can verify safety properties of real-world code. However, using abstract interpretation tools is not always simple. If no user-provided hints are available, the abstract interpretation engine may lose precision during widening and produce an overwhelming number of false alarms. However, manually providing these hints is time consuming and often frustrating when re-running the analysis takes a lot of time. We present an algorithm for program verification that combines abstract interpretation, symbolic execution and crowdsourcing. If verification fails, our procedure suggests likely invariants, or program patches, that provide helpful information to the verification engineer and makes it easier to find the correct specification. By complementing machine learning with well-designed games, we enable program analysis to incorporate human insights that help improve their scalability and usability.

International audience; Credibility, as the general concept covering trustworthiness and expertise, but also quality and reliability, is strongly debated in philosophy, psychology, and sociology, and its adoption in computer science is therefore fraught with difficulties. Yet its importance has grown in the information access community because of two complementing factors: on one hand, it is relatively difficult to precisely point to the source of a piece of information, and on the other hand, complex algorithms, statistical machine learning, artificial intelligence, make decisions on behalf of the users, with little oversight from the users themselves. This survey presents a detailed analysis of existing credibility models from different information seeking research areas, with focus on the Web and its pervasive social component. It shows that there is a very rich body of work pertaining to different aspects and interpretations of credibility, particularly for different types of textual content (e.g., Web sites, blogs, tweets), but also to other modalities (videos, images, audio) and topics (e.g., health care). After an introduction placing credibility in the context of other sciences and relating it to trust, we argue for a quartic decomposition of credibility: Expertise and trustworthiness, well documented in the literature and predominantly related to information source, and quality and reliability, raised to the status of equal partners because the source is often impossible to detect, and predominantly related to the content. The second half of the survey provides the reader with access points to the literature, grouped by research interests. Section 3 reviews general research directions: The factors that contribute to credibility assessment in human consumers of information; the models used to combine these factors; the methods to predict credibility. A smaller section is dedicated to informing users about the credibility learned from the data. Sections 4, 5, and 6 go further into details, with domain-specific credibility, social media credibility, and multimedia credibility, respectively. While each of them is best understood in the context of Sections 1 and 2, they can be read independently of each other. The last section of this survey addresses a topic not commonly considered under "credibility": The credibility of the system itself, independent of the data creators. This is a topic of particular importance in domains where the user is professionally motivated and where there are no concerns about the credibility of the data (e.g., e-discovery and patent search). While there is little explicit work in this direction, we argue that this is an open research direction that is worthy of future exploration. Finally, as an additional help to the reader, an appendix lists the existing test collections that cater specifically to some aspect of credibility. Overall, this review will provide the reader with an organised and comprehensive reference guide to the state of the art and the problems at hand, rather than a final answer to the question of what credibility is for computer science. Even within the relatively limited scope of an exact science, such an answer is not possible for a concept that is itself widely debated in philosophy and social sciences.

Conference of 2014 24th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2014 ; Conference Date: 21 September 2014 Through 24 September 2014; Conference Code:109223; International audience; We propose a Bayesian nonparametrics method, including algorithm for posterior computation, for the sparse regression problem. Our method applies in a general setting, when there are direct or indirect noisy observations of the signal. We try to make a wide focus on smoothness properties and sparsity of the approximate. As an example, we consider the ill-posed inverse problem of Quantum Homodyne Tomography.

Conference of 2014 24th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2014 ; Conference Date: 21 September 2014 Through 24 September 2014; Conference Code:109223; International audience; In order to prevent overfitting in traditional support vector kernel learning, we propose to learn a kernel (jointly with the cost parameter C) by minimizing the single validation estimate with a sequential linear filter algorithm. Additionally, we introduce a simple heuristic in order to improve risk estimation, which randomly swaps several points between the validation and the training sets. Contrarily to previous works, which use several validation sets to improve risk estimation, our strategy does not increase the number of optimization variables. This is easily done thanks to Karasuyama and Takeuchi's multiple incremental decremental support vector learning algorithm. A synthetic signal classification problem underlines the effectiveness of our method. The main parameters of the learned kernel are the finite impulse responses of a filter bank.

[Departement_IRSTEA]Territoires [TR1_IRSTEA]SYNERGIE [Axe_IRSTEA]TETIS-SISOLD4KD'2014: 1st Workshop on Linked Data for Knowledge Discovery with ECML PKDD'2014: The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases; International audience; In this paper we show the potential of contextual itemset mining in the context of Linked Open Data. Contextual itemset mining extracts frequent associations among items considering background information. In the case of Linked Open Data, the background information is represented by an Ontology defined over the data. Each resulting itemset is specific to a particular context and contexts can be related each others following the ontological structure. We use contextual mining on DBpedia data and show how the use of contextual information can refine the itemsets obtained by the knowledge discovery process.

Conference of 22nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN 2014 ; Conference Date: 23 April 2014 Through 25 April 2014; Conference Code:118794; International audience; The Doubly Regularized SVM (DrSVM) is an extension of SVM using a mixture of L2 and L1 norm penalties. This kind of penalty, sometimes referred as the elastic net, allows to perform variable selection while taking into account correlations between variables. Introduced by Wang [1], an efficient algorithm to compute the whole DrSVM solution path has been proposed. Unfortunately, in some cases, this path is discontinuous, and thus not piecewise linear. To solve this problem, we propose here a new sub gradient formulation of the DrSVM problem. This led us to propose an alternative L1 regularization path algorithm. This reformulation effciently addresses the aforementioned problem and makes the initialization step more generic. The results show the validity of our sub-gradient formulation and the efficiency compared to the initial formulation.

International audience; We present a very general algorithm for structured prediction learning that is able to efficiently handle discrete MRFs/CRFs (including both pairwise and higher-order models) so long as they can admit a decomposition into tractable subproblems. At its core, it relies on a dual decomposition principle that has been recently employed in the task of MRF optimization. By properly combining such an approach with a max-margin learning method, the proposed framework manages to reduce the training of a complex high-order MRF to the parallel training of a series of simple slave MRFs that are much easier to handle. This leads to a very efficient and general learning scheme that relies on solid mathematical principles. We thoroughly analyze its theoretical properties, and also show that it can yield learning algorithms of increasing accuracy since it naturally allows a hierarchy of convex relaxations to be used for loss-augmented MAP-MRF inference within a max-margin learning approach. Furthermore, it can be easily adapted to take advantage of the special structure that may be present in a given class of MRFs. We demonstrate the generality and flexibility of our approach by testing it on a variety of scenarios, including training of pairwise and higher-order MRFs, training by using different types of regularizers and/or different types of dissimilarity loss functions, as well as by learning of appropriate models for a variety of vision tasks (including high-order models for compact pose-invariant shape priors, knowledge-based segmentation, image denoising, stereo matching as well as high-order Potts MRFs).

The initial objective of this study, which lies within the field of language teaching andespecially on the role of information and Communication Technology (ICT), is to investigate the potential of ICT in triggering oral language development in the learners of French for Specific purposes (FSP) at Makerere University Business School. This studyadopts action research that focuses on the role of technologies deployed in oral technical presentations of macro-tasks such as the use of MS Office. The aim is to enhance Frenchlearners’ skills in French for Specific purposes. The social constructivist or cultural hypotheses posit that social interaction plays an important role in L2 acquisition (French in this case) in FSP classes through a hybrid environment based on macro-tasks performed indistance and presented in class.The current action research project involved identifying and putting into place a learningsystem for learners of FSP who experienced several difficulties with their spoken French inthe learning process. It further posits that learners construct the new language through socially mediated interaction. Subsequently, this involved establishing whether the use ofPowerPoint presentation (PPP) would engage learners of FSP in collective actions both inthe classroom and in the real world activities. In addition, there was an attempt to establishif relevant web quest materials were likely to enhance oral language acquisition and prompt learners to take responsibility for their own learning., Notre étude relève du domaine de la didactique des langues et plus particulièrement sur la focalisation sur l’utilisation des présentations PowerPoint pour les réalisations des tâches. Ces dernières jouent un rôle important dans le déclenchement de processus d’apprentissage. Lorsque les apprenants réalisent des tâches sur support PowerPoint,interagissent ensemble pour construire du sens, ils sont dotés d’un statut d’acteurs sociaux.Cette étude se donne alors comme objectif d’améliorer la pratique des apprenants en classe et de résoudre certaines de leurs difficultés didactiques et pédagogiques à travers un dispositif hybride. Celui-ci est fondé sur des macro-tâches réalisées majoritairement en semi distance et mise en place majoritairement en présentiel à travers des exposés techniques. Ces derniers ont été conçus pour encourager le développement de performances langagières attendues en production orale des apprenants.Elle se base sur une recherche-action qui a été menée en Ouganda pour des apprenants anglophones qui apprennent le français comme langue étrangère à Makerere UniversityBusiness School (MUBS) à Kampala. Cette étude a porté sur un exposé technique oral parles apprenants du Français sur Objectif Spécifique (FOS) à (MUBS) en Ouganda.Cette recherche-action vise à améliorer l’expression orale en FOS des apprenants de l’université MUBS. Nous avons donc menée des multiples expérimentations qui nous ont permises dans un premier temps de questionner l’impact d’un exposé oral des tâches réelles collectivement élaborée et présentée sur support Power Point sur la production orale de ces apprenants. Ces tâches avaient comme objectif d’encourager les apprenants de transformer un texte en titre, sous-titres et paragraphes, puis de présenter leur travail sous forme de diaporama avec PowerPoint. Le but était d’éviter la lecture linéaire de l’exposé,afin d’accroître leur confiance en eux-mêmes et dans l’interaction en FOS. Grâce à la recherche-action, nous avons pu définir précisément en quoi et pourquoi les choix de nos tâches, de nos activités devraient avoir un effet positif sur notre public cible.

International audience; Multiple kernel learning (MKL) provides flexibility by considering multiple data views and by searching for the best data representation through a combination of kernels. Clinical applications of neuroimaging have seen recent upsurge of the use of multivariate machine learning methods to predict clinical status. However, they usually do not model structured information, such as cerebral spatial and functional networking, which could improve the predictive capacity of the model and which could be more meaningful for further neuroscientific interpretation. In this study, we applied a MKL-based approach to predict prodromal stage of Alzheimer disease (i.e. early phase of the illness) with prior structured knowledges about the brain spatial neighborhood structure and the brain functional circuits linked to cognitve decline of AD. Compared to a set of classical multivariate linear classifiers, each one highlighting specific strategies, the smooth MKL-SVM method (i.e. Lp MKL-SVM) appeared to be the most powerful to distinguish both very mild and mild AD patients from healthy subjets.

International audience; The Artificial Neural Network (ANN) is a black box model capable of resolving paradigms that linear computing cannot. Therefore, the configuration of ANN is a hard task for modeler since it depends on the application complexity. The Discrete EVent system Specification (DEVS) is a formalism to describe discrete event system in a hierarchical and modular way. DEVS is mainly used to defragment system or model in an easy way allowing the interaction with the architecture and behavior of the system. This paper presents a new artificial neural network modeling approach using DEVS formalism in order to facilitate the network configuration by introducing a new scheme of the training phase. We validate our approach with a simple not linearly separable data set example provided by two-dimensional XOR problem.

International audience; This article presents a new approach to individual and collective learning fostered by the use of ICT.From this angle, we aim at going beyond the traditional representation of ICT in education and viewing them instead as necessary instruments in the implementation of community practices of reification of individual or collective knowledge within a socio-technical system.The learning process is then perceived in the light of communicational, social and constructivist criteria in which ICT support the informalization of knowledge, thus contributing to defining a fundamental framework for the emergence of socio-cognitive mediation spaces that become imperative in the construction of knowledge environments.; Cet article expose une nouvelle approche de l'apprentissage individuel et collectif, assisté par l'usage de TIC.Dans cette perspective et dans le cadre d'un système socio-technique, la démarche présentée propose de dépasser la représentation traditionnelle des TIC dans la formation pour les considérer comme des instruments nécessaires à la mise en oeuvre de pratiques communautaires de réification des savoirs individuels ou collectifs.Pour cela, l'activité d'apprentissage est définie selon des dimensions communicationnelles, sociales et constructivistes dans lesquelles les TIC servent de support à « l'informationnalisation des savoirs », contribuant ainsi à la définition d'un cadre essentiel à l'émergence d'espaces de médiations socio-cognitives alors indispensables à la construction d'environnements d'interconnaissances.

International audience; This article presents a new approach to individual and collective learning fostered by the use of ICT.From this angle, we aim at going beyond the traditional representation of ICT in education and viewing them instead as necessary instruments in the implementation of community practices of reification of individual or collective knowledge within a socio-technical system.The learning process is then perceived in the light of communicational, social and constructivist criteria in which ICT support the informalization of knowledge, thus contributing to defining a fundamental framework for the emergence of socio-cognitive mediation spaces that become imperative in the construction of knowledge environments.; Cet article expose une nouvelle approche de l'apprentissage individuel et collectif, assisté par l'usage de TIC.Dans cette perspective et dans le cadre d'un système socio-technique, la démarche présentée propose de dépasser la représentation traditionnelle des TIC dans la formation pour les considérer comme des instruments nécessaires à la mise en oeuvre de pratiques communautaires de réification des savoirs individuels ou collectifs.Pour cela, l'activité d'apprentissage est définie selon des dimensions communicationnelles, sociales et constructivistes dans lesquelles les TIC servent de support à « l'informationnalisation des savoirs », contribuant ainsi à la définition d'un cadre essentiel à l'émergence d'espaces de médiations socio-cognitives alors indispensables à la construction d'environnements d'interconnaissances.

This article presents a new approach to individual and collective learning fostered by the use of ICT.From this angle, we aim at going beyond the traditional representation of ICT in education and viewing them instead as necessary instruments in the implementation of community practices of reification of individual or collective knowledge within a socio-technical system.The learning process is then perceived in the light of communicational, social and constructivist criteria in which ICT support the informalization of knowledge, thus contributing to defining a fundamental framework for the emergence of socio-cognitive mediation spaces that become imperative in the construction of knowledge environments., Cet article expose une nouvelle approche de l'apprentissage individuel et collectif, assisté par l'usage de TIC.Dans cette perspective et dans le cadre d'un système socio-technique, la démarche présentée propose de dépasser la représentation traditionnelle des TIC dans la formation pour les considérer comme des instruments nécessaires à la mise en oeuvre de pratiques communautaires de réification des savoirs individuels ou collectifs.Pour cela, l'activité d'apprentissage est définie selon des dimensions communicationnelles, sociales et constructivistes dans lesquelles les TIC servent de support à « l'informationnalisation des savoirs », contribuant ainsi à la définition d'un cadre essentiel à l'émergence d'espaces de médiations socio-cognitives alors indispensables à la construction d'environnements d'interconnaissances.

International audience; The service composition aims to provide a variety of high level services. Recent approaches cannot fully satisfy the requirement raised by ubiquitous environment. In this paper, we propose a layered design framework which aims at being flexible and robust to failure service composition. It adopts an abstract way of generating plan using rule-based techniques in order to adapt to the changes occurring on the services and the context of use. The approach optimizes the number of services and the recomposition time in large-scale environment by removing the phase of rediscovery. The framework for service composition and monitoring includes learning mechanism for the service selection, based on an estimation of the reputation for abstract services and the quality (QoS) for concrete services. The proposed approach is tested, under USARSim simulator, on a set of ubiquitous services for assisting elderly or dependant person in a residential environment. The obtained results show the feasibility and the scalability of the approach and a better reactivity to the dynamic and uncertain nature of the ubiquitous environment.

International audience; This paper presents a new version of the SVM mixture algorithm initially proposed by Kwok for classification and regression problems. For both cases, a slight modification of the mixture model leads to a standard SVM training problem, to the existence of an exact solution and allows the direct use of well known decomposition and working set selection algorithms. Only the regression case is considered in this paper but classification has been addressed in a very similar way. This method has been successfully applied to engine pollutants emission modeling.

International audience; Conditions complexity of management workshop of production manufacturing increases and requires heuristic algorithm adapted to the context. The principal difficulty then lies in the choice of heuristic algorithm to apply. We propose a help method to schedule, and more specifically a parameter setting up's help of an industrial scheduling software, being based on machine learning system able to extract knowledge from data. An inductive learning based on examples system is developed and replaced in a process of ECD (Extraction of Knowledge starting from Data). The first step pf this process is specific to our problem and uses in this case the capacities of simulation of a market software of scheduling.

International audience; The management of knowledge and know-how becomes more and more important in organizations. Building corporate memories for conserving and sharing knowledge has become a rather common practice. However, we often forget that the efficiency of these activities is strictly connected to the appropriation capacities and learning of the organizational actors. It is through this learning that new skills can be acquired. In this paper, we propose general guidelines facilitating the process of creation and appropriation of professions memories built by means of methods from knowledge engineering and from educational engineering techniques.

Groups of secondary school students conducted guided searches on Internet sites in German. About 120 students were involved in the project. The experiment described in this paper is first used as the basis for a study on the pedagogical relevance of Internet-based research by pupils in foreign languages. We argue that this type of activity is worthwhile only insofar as it promotes the learning process. We present some conclusions on the pitfalls that have to be avoided, on the principles that have to be upheld, and on the new prospects which this medium opens up, in the fields of "active teaching methods", of learner autonomy and of oral and written target-language production. Secondly, we scrutinise the procedure as part of a thorough didactic study, bearing in mind criteria such as relevance, coherence, variation and differentiation. Then we relate the results to studies which have already been published on the subject in France and in Germany. Finally we offer a framework of models showing that our procedure partakes both of a cultural model and of a model sometimes known as the model "of the complex task", and we define these in terms of their dominant characters, the principles that underpin them and their pedagogical objectives., Dans le cadre du cours d'allemand, des classes de lycée (120 élèves concernés) ont effectué des recherches guidées sur des sites Internet. L'expérience décrite sert de base, tout d'abord, à une réflexion sur l'opportunité didactique de recherches documentaires sur Internet, en cours de langue étrangère. En effet, l'intérêt de cette activité n'est réel que si elle donne lieu à la mise en place d'un dispositif d'apprentissage. Des conclusions sont tirées sur les écueils à éviter, les principes à respecter et sur les perspectives nouvelles que ce média ouvre dans le domaine des "méthodes actives", de l'autonomisation de l'apprenant et de la production en langue étrangère. Le dispositif utilisé fait ensuite l'objet d'une réflexion didactique approfondie. Il est passé au crible des critères de variation, de différenciation, de pertinence et de cohérence. Puis les résultats sont comparés aux travaux publiés dans ce domaine en France et en Allemagne. Enfin, une proposition de modélisation fait apparaître que le dispositif utilisé a une double appartenance : au modèle culturel et au modèle de la tâche complexe, ceux-ci étant définis par leurs dominantes, leurs activités typiques, leurs objectifs formatifs, leurs principes didactiques et leurs finalités éducatives.