Frédéric Blanchard

A new pixel-oriented visualization technique through color image

Color image is often considered as a fundamental perceptual unit of visualization. In this paper, we suggest using this medium (color image) to summarize multidimensional data and thus to turn a data set into a meaningful insight. The methodology we use is based on the theory of Keim for designing pixel-oriented visualization techniques. The technique we propose consists in a three-step pipeline. The first one is devoted to dimensionality reduction by projecting multidimensional data into a three-dimensional space. In this work, we use the classical principal component analysis (PCA) to reduce the dimension to three. The second step, called color mapping, is based on the reverse color information transformation defined by Ohta et al. This stage is the main novelty of this work in addition to the pipeline itself. The third step consists in a pixel-oriented method to display large data sets with an image using space-filling curve techniques. The combination of these steps (first, dimensionality reduction with PCA, second, color mapping with color information of Ohta et al., and third, space-filling curve with Peano–Hilbert curve) allows us to obtain a new unsupervised visualization technique through color images. This blind (i.e. unsupervised) technique using a color image gives a previsualization that can be used before exploring the data set or choosing more effective colors. Some applications are proposed in the field of multicomponent image visualization.

Energy and activity monitoring over wireless sensor networks

Wireless sensor networks are one the most challenging fields in computer science and networking in particular with the exposition of Internet of things. The energy management is a crucial issue on wireless sensor network since it has a real influence on then network lifetime. Since such networks are used for many sensitive applications, they need to guarantee some quality of service requirements as continuous connectivity or continuous monitoring or balanced energy consumption over all the network nodes. In this paper, we suggest an algorithm which monitors the energy consumption of all nodes (in relation with their main activity: communication). We present an activity model of the network and evaluate how the remaining energy behaves. Our main objective is to predict (after a learning step) the energy consumption of all the nodes. Then, various techniques could be used to find a cure to such troubles (wake up sleeping neighbor nodes, increasing radio ranges of some neighbor nodes, etc). Simulations have shown interesting results, and large‐scale evaluation has to be undertaken to confirm the early results. Copyright

New similarity index based on the aggregation of membership functions through OWA operator

In the field of data analysis, the use of metrics is a classical way to assess pairwise similarity. Unfortunately the popular distances are often inoperative because of the noise, the multidimensionality and the heterogeneous nature of data. These drawbacks lead us to propose a similarity index based on fuzzy set theory. Each object of the dataset is described with the vector of its fuzzy attributes. Thanks to aggregation operators, the object is fuzzified by using the fuzzy attributes. Thus each object becomes a fuzzy subset within the dataset. The similarity of a reference object compared to another one is assessed through the membership function of the fuzzified reference object and an aggregation method using OWA operator.

Link-gain based power control mechanism for Wireless Sensor Networks

Recently, the topology control strategies attract more and more the research community as being a crucial issue to be handled for the Wireless sensors networks (WSNs). In this paper, we propose a centralized power control mechanism based on the link gain information to reduce network energy consumption. Our mechanism works through two steps, first we construct a skeleton that ensure network connectivity while reducing considerably the number of links in the network. Second, based on the obtained skeleton, we reduce the power transmission for each node while keeping a connectivity with a node of the skeleton with a link gain greater than a predefined threshold. The simulation results conducted over TOSSIM, show that our proposal can save the network energy up to 10% while achieving better PDR value, which goes near to 12% compared to non-optimized network.

A new way for the exploration of a dataset based on a social choice inspired approach

The exploration of a data set consists in grouping similar data. The classical statistical methods often fail when there is is no minimal assumption on the clusters. Our approach is based on the links between data, but the pairwise comparison between data and the importance of the links depend heavily on context where data lies. We propose to analyze a dataset through methods of the social choice theory where data plays both the role of a candidate and the role of a voter. The candidates are ranked by the voters and each voter gives a score to each candidate according to his ranking. We propose one specific election for each voter based on his preferences. The voters of these elections have weights computed according to their respective behaviors. In this approach, the conventional similarity indices between data are used to define the electoral behavior of each data.

Centralized energy monitoring over wireless sensor networks

Wireless sensor networks (WSN) are one the most challenging field in computer and networking science. The energy management is a crucial issue on WSN since it has a real influence on a network lifetime. Since such networks are used for many sensitive applications, they may need to guarantee some Quality of Service requirements as continuous connectivity or continuous monitoring or balanced energy consumption over all network nodes. In this paper, we suggest a centralized approach which monitors the energy consumption of all nodes (in relation with their main activity: communication). We present an activity model of the network and evaluate how the remaining energy behaves. Our main objective is to predict (after a learning step) the energy consumption of all the nodes. Many techniques could be used to find a cure to such troubles (wake up sleeping neighbor nodes, increasing radio ranges of some neighbor nodes,. . . ).

Structuring complex data using representativeness graphs

This contribution addresses the problem of extracting some representative data from complex datasets and connecting them in a directed forest. First we define a degree of representativeness (DoR) based on the Borda aggregation procedure. Secondly we present a method to connect pairwise data using neighborhoods and the DoR as an objective function. We then present three case studies as a proof of concept: unsupervised grouping of binary images, analysis of co-authorships in a research team and structuration of a medical patient-oriented database for a case-based reasoning use.

Singular profile of diabetics

The therapeutic monitoring of patients at home produces a mass of data that requires new methods for analyzing and processing. The main challenge of medical data processing is the management of high intra-subject and inter-subject variabilities. The need for specific dashboards for both the patient and the group of patients with similar therapeutic behaviors is another difficulty. This paper describes a new way to analyze such medical data through the use of singular profiles of elderly patients in a population with type 2 diabetes. Our goal is to develop a methodology of data processing for following the insulin therapy at home. The first step of processing consists in the fuzzification of the attributes within the data samples to ensure the robustness of the method. The singularity index we propose assesses the fuzzy attributes relative to each patient. This index is obtained by computing the power of the fuzzy set associated with each attribute. The singularity of the attributes permits us to give the singular profile of each patient. The visualization step leads us to propose empirical rules to obtain three kinds of different profiles. This robust approach also permits us to highlight three clusters of elderly diabetics. The three clusters appear very similar as the ones obtained when using classical automated methods of clustering such as the k-medoids. By extending this approach, the ultimate goal of our future developments is the design of a recommender system for type 2 diabetics with insulin therapy.