Search

This paper describes an efficient combinatorial method for simplification of topological features in a 3D scalar function. The Morse-Smale complex, which provides a succinct representation of a function's associated gradient flow field, is used to identify topological features and their significance. The simplification process, guided by the Morse-Smale complex, proceeds by repeatedly applying two atomic operations that each remove a pair of critical points from the complex. Efficient storage of the complex results in execution of these atomic operations at interactive rates. Visualization of the simplified complex shows that the simplification preserves significant topological features while removing small features and noise.

This paper describes a highly flexible component architecture, primarily designed for automotive control systems, that supports distributed dynamically- configurable context-aware behaviour. The architecture enforces a separation of design-time and run-time concerns, enabling almost all decisions concerning runtime composition and adaptation to be deferred beyond deployment. Dynamic context management contributes to flexibility. The architecture is extensible, and can embed potentially many different self-management decision technologies simultaneously. The mechanism that implements the run-time configuration has been designed to be very robust, automatically and silently handling problems arising from the evaluation of self- management logic and ensuring that in the worst case the dynamic aspects of the system collapse down to static behavior in totally predictable ways.

We present visibility computation and data organization algorithms that enable high-fidelity walkthroughs of large 3D geometric data sets. A novel feature of our walkthrough system is that it performs work proportional only to the required detail in visible geometry at the rendering time. To accomplish this, we use a precomputation phase that efficiently generates per cell vLOD: the geometry visible from a view-region at the right level of detail. We encode changes between neighboring cells' vLODs, which are not required to be memory resident. At the rendering time, we incrementally construct the vLOD for the current view-cell and render it. We have a small CPU and memory requirement for rendering and are able to display models with tens of millions of polygons at interactive frame rates with less than one pixel screen-space deviation and accurate visibility.

Artificial neural networks have been extensively applied to document analysis and recognition. Most efforts have been devoted to the recognition of isolated handwritten and printed characters with widely recognized successful results. However, many other document processing tasks, like preprocessing, layout analysis, character segmentation, word recognition, and signature verification, have been effectively faced with very promising results. This paper surveys the most significant problems in the area of offline document image processing, where connectionist-based approaches have been applied. Similarities and differences between approaches belonging to different categories are discussed. A particular emphasis is given on the crucial role of prior knowledge for the conception of both appropriate architectures and learning algorithms. Finally, the paper provides a critical analysts on the reviewed approaches and depicts the most promising research guidelines in the field. In particular, a second generation of connectionist-based models are foreseen which are based on appropriate graphical representations of the learning environment.

ID: 1018

The formal approach to visual language definition is to use graph grammars and/or graph transformation techniques. These techniques focus on specifying the syntax and manipulation rules of the concrete representation. This paper presents a constraint and object-oriented approach to defining visual languages that uses UML and OCL as a definition language. Visual language definitions specify a mapping between concrete and abstract models of possible visual sentences, which carl subsequently be used to determine if instances of each model "validly" express each other. This technique supports many:many mappings between concrete and abstract model instances, and supports the implementation of functionality that requires feedback from the abstract domain to the concrete.

In this paper we describe a tools environment which automates the validation and maintenance of a requirements model written in many-sorted first order logic. We focus on: a translator, that produces an executable form of the model; blame assignment functions, which input batches of\ud mis-classified tests (i.e. training examples) and output likely faulty parts of the model; and a theory reviser, which inputs the faulty parts and examples and outputs suggested revisions to the model. In particular, we concentrate on the problems encountered when applying these tools to a real application: a requirements model containing air traffic control separation standards, operating methods and airspace information.

Contrastive representation learning has proven to be an effective self-supervised learning method. Most successful approaches are based on Noise Contrastive Estimation (NCE) and use different views of an instance as positives that should be contrasted with other instances, called negatives, that are considered as noise. However, several instances in a dataset are drawn from the same distribution and share underlying semantic information. A good data representation should contain relations, or semantic similarity, between the instances. Contrastive learning implicitly learns relations but considering all negatives as noise harms the quality of the learned relations. To circumvent this issue, we propose a novel formulation of contrastive learning using semantic similarity between instances called Similarity Contrastive Estimation (SCE). Our training objective is a soft contrastive learning one. Instead of hard classifying positives and negatives, we estimate from one view of a batch a continuous distribution to push or pull instances based on their semantic similarities. This target similarity distribution is sharpened to eliminate noisy relations. The model predicts for each instance, from another view, the target distribution while contrasting its positive with negatives. Experimental results show that SCE is Top-1 on the ImageNet linear evaluation protocol at 100 pretraining epochs with 72.1% accuracy and is competitive with state-of-the-art algorithms by reaching 75.4% for 200 epochs with multi-crop. We also show that SCE is able to generalize to several tasks. Source code is available here: https://github.com/CEA-LIST/SCE., Comment: Accepted to IEEE Winter Conference on Applications of Computer Vision (WACV) 2023

Preprint; International audience; The design of immersive interaction for mixed reality based on headmounted displays (HMDs), hereafter referred to as Mixed Reality (MR), is still a tedious task which can hinder the advent of such devices. Indeed, the effects of the interface design on task performance are difficult to anticipate during the design phase: the spatial layout of virtual objects and the interaction techniques used to select those objects can have an impact on task completion time. Besides, testing such interfaces with users in controlled experiments requires considerable time and efforts. To overcome this problem, predictive models, such as the Keystroke-Level Model (KLM), can be used to predict the time required to complete an interactive task at an early stage of the design process. However, so far these models have not been properly extended to address the specific interaction techniques of MR environments. In this paper we propose an extension of the KLM model to interaction performed in MR. First, we propose new operators and experimentally determine the unit times for each of them with a HoloLens v1. Then, we perform experiments based on realistic interaction scenarios to consolidate our model. These experiments confirm the validity of our extension of KLM to predict interaction time in mixed reality environments.

International audience; The provision of public data through various Database-as-a-Service (DBaaS) providers has recently emerged as a significant trend, backed by major organisations. This paper introduces Nebula, a non-profit middleware providing multicloud querying capabilities by fully outsourcing its users' queries to the involved DBaaS providers. First, we propose a quoting procedure for those queries, whose need stems from the pay-perquery policy of the providers. Those quotations contain monetary cost and response time estimations, and are computed using provider-generated tenders. Then, we present an agent-based dynamic optimisation engine that orchestrates the outsourced execution of the queries. Agents within this engine cooperate in order to meet the quoted values. We evaluated Nebula over simulated providers by using the Join Order Benchmark (JOB). Experimental results showed Nebula's approach is, in most cases, more competitive in terms of monetary cost and response time than existing work in the multi-cloud DBMS literature.

The movement of data between processing and memory units, often referred to as the ‘von Neumann bottleneck’ is the main reason for the degraded performance of contemporary computing systems. In an effort to overcome this bottleneck, methods to ‘compute’ at the location of data are being pursued in many emerging memories, including Resistive RAM (ReRAM). Although many prior works have pursued addition in memory, the latency of n-bit addition has not been judiciously optimized, resulting in O(n) or at best O(log(n)). Computing with three states can enable carry-free addition and result in a latency which is independent of operand width (O(1)). In this work, we propose a method to perform carry-free addition completely in memory (a storage array, a processing array and their peripheral circuitry). The proposed technique incurs a latency of 22 memory cycles, which outperforms other in-memory binary adders for n > 32. This speed is achieved at the cost of increased peripheral hardware.

International audience; Conversational information seeking is a major emerging research area because of the increasing popularity of conversational AI systems users utilize to perform their search tasks. Search systems and multiple other user supporting applications benefit from modeling the search tasks users carry out to satisfy their information needs. Most existing search task modeling methods are monolingual, and few methods leverage user clicks even though clicked URLs are crucial for modeling user intent. We propose a language-agnostic, user intent aware approach to model search tasks from user interactions with search systems. The proposed approach leverages user intent modeling from clicked query-document pairs, latent representations of queries in a language-agnostic space, and graph-based clusteringto model search tasks in an unsupervised approach. Experimental results demonstrate the proposed approach outperforms recent work in search task modeling, supporting user queries in multiple languages. It can also produce search task modeling results in the order of milliseconds, an essential aspect for conversational systems and user support applications requiring realtime results.

International audience; The routing is a real-world engineering problem in IoT Collection Networks. It has a direct impact on the network performance. The routing problem can be treated as a multi-objective optimization problem. Our objective is then to study the IoT routing problem using the Multi-Objective Grey Wolf Optimizer (MOGWO) [1], which consists in translating the routing process into a multi-objective optimization problem and implementing its execution with the MOGWO. In this context, the experimental study showed that the MOGWO provides better quality of routing (higher lifetime of the network, more efficient delivery delay and higher number of neighbors).

International audience; The sensor deployment problem consists in finding an optimal (or near-optimal) way of placing sensors in a given area in order to cover the whole space. However, sometimes it is not necessary to monitor the complete area; one may want to focus on some points considered as target. This leads to the target coverage problem. This problem includes two main processes: (1) placing sensors around the targets, (2) scheduling the activation time redundant sensors in order to lengthen the network lifetime. This paper proposes a multi-objective approach based on genetic algorithms that aims to simultaneously find good positions of sensor nodes and the maximum number of disjoint cover sets. A new chromosome (solution) encoding in which the genes contain both, the position and the identifier of sensor owning group is introduced. The efficiency of the proposed approach is assessed by a comparison with existing methods.

International audience; Objective evaluation of community detection algorithms is a strategic issue. Indeed, we need to verify that the communities identified are actually the good ones. Moreover, it is necessary to compare results between two distinct algorithms to determine which is most effective. Classically, validations rely on clustering comparison measures or on quality metrics. Although, various traditional performance measures are used extensively. It appears very clearly that they cannot distinguish community structures with different topological properties. It is therefore necessary to propose an alternative methodology more sensitive to the community structure variations in order to conduct more effective comparisons. In this paper, we present a framework to tackle this challenge through a comprehensive analysis of the community structure of overlapping community structured networks. We illustrate our approach with an experimental analysis of a real-world network with a ground-truth community structure that we compare with the output of eight different overlapping community detection procedures, representative of categories of popular algorithms available in the literature. The results allow a better understanding of their behavior. Furthermore, they demonstrate that more emphasis should be put on the topology of the uncovered community structure in order to evaluate the effectiveness of community detection algorithms.

International audience; In this paper, we develop and evaluate a robust single-lead electrocardiogram (ECG) delineation system and waves localization method based on nonlinear filtering approach. This system is built in two phases. In the first phase, we proposed a mathematical model for detecting ECG features like QRS complex peak, P and T-waves onsets and ends from noise free of synthetic ECG signal. Later, we develop a theoretical model to obtain real approach for detecting these features from real noisy ECG signals. To evaluate our method, it has been applied on electrocardiogram signals of QT-MIT standard database, the QRS peak detection obtained sensitivity (Se) 98.88 and a positive productivity P+ =98.43. For P-onset, P-end, T-end evaluations, this approach provides Sensitivity (Se) of 75.16, 71, and 90.7 respectively. Another test has been conducted to evaluate the performance of this approach against different levels of noise with the same synthetic ECG signal. For denoising test on electrocardiogram signals, our approach based on nonlinear filtering scheme (NLFS) gives robust performance against noise.

International audience; Influence in Twitter has become recently a hot research topic since this micro-blogging service is widely used to share and disseminate information. Some users are more able than others to influence and persuade peers. Thus, studying most influential users leads to reach a large-scale information diffusion area, something very useful in marketing or political campaigns. In this paper, we propose a new approach for influence assessment on Twitter network, it is based on a modified version of the conjunctive combination rule in belief functions theory in order to combine different influence markers such as retweets, mentions and replies. We experiment the proposed method on a large amount of data gathered from Twitter in the context of the European Elections 2014 and deduce top influential candidates.

International audience; Various dynasties ruled the Indian sub-continent and left behind enormous and rich cultural heritage that also included intellectually enriched research in the shape of various documents scripted in Urdu. In order to provide efficient access to this knowledge, analysis though digitizing the existing work is the need of hour. In addition to digitization, efficient search mechanisms also need to be implemented to provide users a rapid access to the queried information. In most cases, the digitized documents are complemented by manually assigned tags which not only is a time consuming task but also provides a very limited search facility. Automating the transcription of these documents using Optical Character Recognition (OCR) systems is also challenging due to the very complex cursive nature of Urdu text. To overcome these limitations, a keyword spotting based information retrieval system for document images is introduced in this study. The proposed technique relies on two major modules, document indexing and retrieval. Images of documents are segmented into partial words (ligatures) and identical partial words (PWs) are grouped into clusters. We introduce the concept of considering each (partial) word as a unique shape and a set of shape descriptors is extracted to characterize the PWs. The clusters of PWs are used to index a given set of documents. During retrieval, the query word presented to the system is matched with the clusters in the database and all documents containing instances of the query word are retrieved and presented to the user. The system evaluated on a set of printed Urdu documents in Nastaliq font realized promising precision and recall rates.

International audience; During their geometry processing and transmission 3D meshes are subject to various visual processing operations like compression, watermarking, remeshing, noise addition and so forth. In this context it is indispensable to evaluate the quality of the distorted mesh, we talk here about the mesh visual quality (MVQ) assessment. Several works have tried to evaluate the MVQ using simple geometric measures, However this metrics do not correlate well with the subjective score since they fail to reflect the perceived quality. In this paper we propose a new objective metric to evaluate the visual quality between a mesh with a perfect quality called reference mesh and its distorted version. The proposed metric uses a chosen statistical distribution to extract parameters of two random variable sets, the first set is the dihedral angles related to the reference mesh, while the second set is the dihedral angles related to the distorted mesh. The perceptual distance between two meshes is computed as the Kullback-Leibler divergence between the two sets of variables. Experimental results from two subjective databases (LIRIS masking database and LIRIS/EPFL general purpose database) and comparisons with seven objective metrics cited in the state-of-the-art demonstrate the efficacy of the proposed metric in terms of the correlation to the mean opinion scores across these databases.

International audience; For over twenty years the EDOC conference has been the primary annual event for disseminating and discussing the latest developments in the area of enterprise computing. The workshop program is an important satellite event of the EDOC conference. Workshops cover more focused topics and allow for the presentation and discussion of work that is in the earlier development stages. As such, the workshops provide an excellent forum for discussing topics from the area of enterprise computing that have the potential to become important research streams within the next few years, and for discussing topics that are already important in a smaller and more focused setting. In addition to the workshop program, the EDOC Demonstration track offers an exciting and highly interactive outlet for researchers and practitioners to present prototypes and applications in the context of enterprise computing. Four demonstrations are presented this year. 2020 is the year of the COVID-19 pandemic, with a strong impact on research and social life. Despite the challenges of new organization (the conference and the workshops will be held on-line), we are proud to present an interesting program and the four workshops that will be held in conjunction with EDOC this year:• The workshop on Trends in Enterprise Architecture Research, (TEAR)• The workshop on Service-oriented Enterprise Architecture for Enterprise Engineering (SoEA4EE)• The workshop on Privacy and Security in Enterprise Modeling (PriSEM)• The workshop on Frontiers of Process Aware Systems (FoPAS).[...]

International audience; BACKGROUND AND MOTIVATION: The world population is living longer than ever before, Ambient Assisted Living (AAL) research try to ensure people are ageing well by using digital and physical technologies. The development of AAL platforms presents several challenges (e.g., sensor data integration, device interoperability). Many requirements for AAL platforms can be met by using the Internet of Things (IoT) and Cloud Computing paradigm standards. METHODOLOGY: In this paper, we conducted a systematic mapping of the literature on AAL platforms to investigate the correlation between these three paradigms. We focused on AAL platforms that use IoT and Cloud Computing concepts to meet their requirements. Our primary research question is “What are the requirements of AAL platforms in the IoT Era and their relationship with IoT standards?”. We defined as the scope of our systematic mapping the primary studies about AAL platforms published between 2013 to 2017, and we used seven indexed electronic databases. RESULTS: We obtained 35 papers whose AAL platforms developed using IoT or Cloud Computing standards. We then analyse and correlate the requirements, architectures, and evaluation types presented by these platforms. We found patterns by connecting specific requirements with platform design. Webelieve these results assist AAL developers and architectural designers in the creation of new AAL platforms and applications.

International audience; Our regulatory systems have attempted to keep abreast of new technologies by recalibrating and adapting our regulatory frameworks to provide for new opportunities and risks, to confer rights and duties, safety and liability frameworks, and to ensure legal certainty for businesses. These adaptations have been reactive and sometimes piecemeal, often with artificial delineation on rights and responsibilities and with unintended flow-on consequences. Previously, technologies have been deployed more like tools, but as autonomy and self-learning capabilities increase, robots and intelligent AI systems will feel less and less like machines and tools. There is now a significant difference, because machine learning AI systems have the ability to learn, adapt their performances and ‘make decisions’ from data and ‘life experiences’. This chapter provides brief insights on some of the topical developments in our regulatory systems and the current debates on some of the risks and challenges from the use and actions of AI, autonomous and intelligent systems [1].

International audience; This chapter focuses on the ethics of narrow, as opposed to general AI. It makes the practical as well as the philosophical case for discussion of AI ethics. It considers ethical charters, then discusses the principal ethical issues: bias, explainability, liability for failure, harmlessness, the ethical use of data, whether AIs should have legal personality, the effects on employment and society, and AIs impersonating humans. A case study is presented of AI in personal insurance. It makes the case for regulation of AI and discusses the challenges of enacting regulation. It draws conclusions, that the benefits of AI are so valuable that the ethical risks must be managed, or the benefits may be lost because of the loss of public trust. There are grounds for optimism, notably the public consciousness of the issues, the engagement of governments and the amount of private and public investment in ethical research.

International audience; Developing products for improving plant nutritional status, e.g. fertilizers or plant growth regulators, is an important topic to move towards sustainability in agriculture and to ensure to feed the world population. A key challenge is to identify when, what, and how much nutrients to add to plants' growth environment. In this paper, we study a use case on how to characterize rapeseed plant nutrient deficiencies during their growth. A promising approach consists of deriving data from spectroscopy of leaves, and using this representation to predict what kind of deficiency (if any) plants are undergoing. We are considering three research questions: 1) from which day after onset of a nutrient deficiency we can identify it, 2) whether leaves that have sprouted under nutrient-rich conditions can still help in identifying problems. Third, and most importantly, performing the spectroscopy on the full range of wavelengths is expensive, which under production conditions allows for only relatively few samples. We therefore explore how to perform dimensionality reduction and preprocessing to achieve good predictive accuracy. We show that 1) deficiencies can be identified early on, 2) leaf generations help to predict nutrient deficiencies, and 3) that preprocessing increases the accuracy and dimensionality reduction can be performed without loss of accuracy. Along the way, we find that our some of our industry partners' assumptions about the data do not seem to be borne out by our empirical results, and that the subset of data they initially selected turns out to be too easy to model. The full data leads to more informative insights.

The goal of this work is to provide a brief classification of some Shortest Path Problem (SPP) variants that include edge constraints and that find applications in several different contexts, including optical networks, transportation and logistics. One of the most broad and notable classes of edge-constrained SPPs is given by Resource Constrained Shortest Path Problems (RCSPPs). In RCSPPs, in addition to the customary directed graph $G=(V, A)$ and edge-distance function, a L-dimensional vector of resources R is defined. Essentially, each resource is related to relevant link attributes that need to be accounted for in the planning of the path. Accordingly, a path P* is optimal whenever it is minimal w.r.t. the distance function, and satisfies the restrictions enforced on the resources Other variants can involve colors assigned to the nodes and/or the edges of the graph and/or reload costs. These kinds of problems have relevant applications in network reliability. Particularly interesting is the so-called k-Color SPP, where a color is assigned to each edge and the minimum path from a given source node s to a target node t must not cross more than k differently colored edges. As for the use of reload cost, a reload cost r(b, c) is assigned to each pair of colors (b, c) and represents the amount to be paid if in a path P an arc of color c is traversed after an arc of color b.

International audience; Internet of Things (IoT) is a network where several and various interconnected, smart objects continuously generate alarge amount of data transmitted over Internet. Several efforts and research works in IoT focused on hardware constraints, lowcost, application design, as well as good communication technologies. However, due to the intrinsic characteristics of diversity,heterogeneity, large-scale, dynamic and large volume of data generated and various IoT applications needs, using traditionaldatabase management mechanisms and analytics architectures is not generally suitable. Then, it is challenging to provideefficient IoT data storage and query processing mechanisms for satisfying IoT application needs. This paper identifies the mainspecifications and mechanisms of data and query management for IoT and classifies them. Moreover, this work presentsadvances on data and query management mechanisms on IoT, shows their advantages and limits and discusses the challengingopen research issues that need to be addressed for providing guidelines for further contributions.

International audience; In this paper, we propose a low-cost n-dimensional cache (nD-Cache) architecture for FPGA-Based image and signal processing Systems On Chip (SoCs). The architecture allows efficient access to structured data as in 2D or 3D images. We developed a theoretical model for our architecture. It gives a methodology to define cache's practical implementation based on the application and system parameters. Complexity and performance measurements for selected image processing algorithms like jumping snake and 2D Back-Projection are done and compared to classical solutions like associative caches. The architecture is shown to be efficient for tracking algorithm applications by exploiting spacial and temporal locality. Numerical results indicate that 50% improvement in run-time performance can be achieved.

International audience; In this paper, we present the problem of multivariate function decompositions into sums and compositions of monovariate functions. We recall that such a decomposition exists in the Kolmogorov's superposition theorem, and we present two of the most recent constructive algorithms of these monovariate functions. We first present the algorithm proposed by Sprecher, then the algorithm proposed by Igelnik, and we present several results of decomposition for gray level images. Our goal is to adapt and apply the superposition theorem to image processing, i.e. to decompose an image into simpler functions using Kolmogorov superpositions. We synthetise our observations, before presenting several research perspectives.

Dual-frequency ultrasonic transducers, capable of operating over distinct frequency bands, may be conveniently used in different applications such as harmonic imaging. This paper presents the design and fabrication of a dual-frequency hybrid transducer prototype, consisting of a Capacitive Micromachined Ultrasound Transducer (CMUT) placed on top of a piezoelectric transducer stack. The low-frequency (LF) transducer, i.e. a PZT disk supported by a backing and matched using a quarter-wavelength layer, was designed for transmit (TX) and receive (RX) operation around 7 MHz in immersion, while the high-frequency (HF) transducer, i.e. a single-element Re-verse-Fabricated CMUT, was designed for RX operation around 14 MHz. Finite Element Modeling (FEM) was used to investigate the behavior of the proposed hybrid transducer stack. A hybrid transducer prototype was then fabricated. Two additional transducer prototypes, i.e. a single-element piezoelectric and a single-element CMUT, were fabricated using the same materials. The three transducer prototypes were electrically and acoustically characterized. Characterization results were compared with simulations showing that the proposed hybrid transducer approach can be used to increase the RX frequency bandwidth, thus demonstrating its potential in nonlinear imaging applications.

Biplane modality is frequently used in medical ultrasound imaging applications. Crossed array configurations, in which two orthogonal 1-D arrays co-exist on the same area, are particularly suitable for biplane imaging. However, the imaging performance is limited by the absence of focusing capability in the arrays respective elevation directions. In this paper, we present a novel 120+120-element 7 MHz crisscross Capacitive Micromachined Ultrasonic Transducer (CMUT) probe in which the orthogonal arrays are both provided with Fresnel elevation focusing capability. The CMUT probe includes low-noise amplifiers and high-voltage switches, which allow fast switching (

© 2018 IEEE. MVMS-based Capacitive Micromachined Ultrasonic Transducer (CMUT) technology enables the development of advanced transducer-electronics integrated multi-chip modules (MCM) for medical imaging applications. In this pajer, we present the design, fabrication and characterization of a 256-element CMUT spiral array integrated with a 256-channel analog front end (AFE) ASIC featuring high-voltage unipolar pulsers, low-noise amplifiers and a programmable transmit beamformer. The spiral array, designed to perform volumetric beam steering at 7 MHz, is characterized by a c density-tapered layout of 220μm-wide hexagonal elements distributed over a quasi-circular area of 10 mm diameter. The CMUT and the AFE were co-designed, fabricated and interconnected through an acoustically optimized 3-D packaging method. The resulting MCM was electromechanically and acoustically characterized, demonstrating the potential of the proposed approach for integrated 2-D CMUT array fabrication. ispartof: pages:1-4 ispartof: 2018 IEEE International Ultrasonics Symposium (IUS) vol:2018-January pages:1-4 ispartof: 2018 IEEE International Ultrasonics Symposium (IUS) location:Kobe date:22 Oct - 25 Oct 2018 status: Published online

Hashtags are crucial social data that are useful for multiple purposes, such as tweets' classification, indexing, content categorization, search or recommendation. Hashtags recommendation on Twitter is an important issue, seeing the importance of this social data and the need of social users to be provided with personalized information that fits their interests. Users recommendation is also an eminent task as it allows a user to broaden his network by users with similar interests. This work focuses on how hashtags can be analyzed for recommendation purposes. First, we present a state of the art reviewing social analysis, users and hashtags' recommendation. We then present a new architecture of hashtags and users recommender system based on hashtags' semantic analysis. This proposal is based on ontology as a semantic resource. We build social user profiles, analyze hashtags and study their contextual and temporal cooccurrence. We then propose two novel ranking schemes. The first is HF-IUTF scheme that weights hashtags in the user profile. We use it to rank and filter the representative hashtags of each user's social profile. The second contribution is HF-IGHF scheme: Given a group of users, this method identifies the hashtags representative of the group. We apply spectral clustering algorithm to obtain similar clusters. We can then recommend similar users of the same cluster and representative new and recent hashtags of users in the cluster. The originality of this architecture is the way user profiles will be semantically indexed and analyzed, and ranking schemes that are independent of the way the similarity group is built.

International audience; VR has now come from industry to everyday application. Mainstream software and devices allow artists to create contents with a fast learning curve. Since 2014, with the launch of Google Cardboard and 360 cameras at a reasonable price, with the massive success of Unity 3D and Unreal UDK, real-time immersion no longer stands in the hands of experts but spreads to creative enthusiasts which result in a huge production of content. Like at the early age of photography and then cinema, slowly emerge questions about composition, narrative structure and visual grammar. This article is a raw presentation of issues of narrative grammar in 360.

International audience; Nowadays, more and more sources (connected devices, social networks, etc.) emit real-time data with fluctuating rates over time. Existing distributed stream processing engines (SPE) have to resolve a difficult problem: deliver results satisfying end-users in terms of quality and latency without over-consuming resources. This paper focuses on parallelization of operators to adapt their throughput to their input rate. We suggest an approach which prevents operator congestion in order to limit degradation of results quality. This approach relies on an automatic and dynamic adaptation of resource consumption for each continuous query. This solution takes advantage of i) a metric estimating the activity level of operators in the near future ii) the AUTOSCALE approach which evaluates the need to modify parallelism degrees at local and global scope iii) an integration into the Apache Storm solution. We show performance tests comparing our approach to the native solution of this SPE.

A novel fully wearable system based on a smart wristband equipped with stretchable strain gauge sensors and readout electronics have been assembled and tested to detect a set of movements of a hand crucial in rehabilitation procedures. The high sensitivity of the active devices embedded on the wristband do not need a direct contact with the skin, thus maximizing the comfort on the arm of the tester. The gestures done with the device have been auto-labeled by comparing the signals detected in real-Time by the sensors with a commercial infrared device (Leap motion). Finally, the system has been evaluated with two machine-learning algorithms Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM), reaching a reproducibility of 98% and 94%, respectively.

International audience; Projector-camera systems, used in Spatial AugmentedReality, automatically adapt the video projections tothe scene objects according to the visualization conditions.This paper introduces a novel non-invasive (without StructuredLight) method based on a combination of traditionalFeature Matching (FM) and more computationally effectiveOptical Flow (OF). It requires only one projected and oneacquired image at a time in the most difficult case when bothprojected content and geometric transformations changeevery frame. It detects scene changes when OF fails andthus should be replaced by FM. In the experiments, weshow that the method yields a more precise and less shakycompensation for different types of projected videos, and isup to 2.8 times faster than previous FM-based works.

International audience; Part of the research effort in automatic person discovery in multimedia content consists in analyzing the errors made by algorithms. However exploring the space of models relating algorithmic errors in person discovery to intrinsic properties of associated shots (e.g. person facing the camera) - coined as post-hoc analysis in this paper - requires data cura- tion and statistical model tuning, which can be cumbersome. In this paper we present a visual and interactive tool that facilitates this exploration. Adequate statistical building blocks are defined, and coordinated by visual and interactive components inspired from the literature in information visualization. A case study is conducted with multimedia researchers to validate the tool. Real data obtained from the MediaEval person discovery task was used for this experiment. Our approach yielded novel insight that was completely unsuspected previously.

While near threshold design space provides a promising approach towards energy-efficient computing, it is plagued by sub-optimal performance. Application characteristics and hardware non-idealities of conventional architectures (optimized for the nominal voltage) prevent us from fully leveraging the potential of NTC systems. Further, the popular approach of increasing the computational core count to compensate for the performance loss severely burdens the on-chip communication fabric with an increased communication demand. In this work, we quantitatively analyze the performance bottleneck created by a conventional NoC architecture in many-core NTC systems. To reclaim the performance lost due to a sub-optimal NoC, we propose BoostNoC — a power efficient, multi-layered network-on-chip architecture. BoostNoC improves the system performance by nearly 2× over a conventional NTC system. Further, we improve the energy efficiency by 1.4× with the use of drowsy routers.