Search

Feature selection is the process of choosing, or removing, features to obtain the most informative feature subset of minimal size. Such subsets are used to improve performance of machine learning algorithms and enable human understanding of the results. Approaches to feature selection in literature exploit several optimization algorithms. Multi-objective methods also have been proposed, minimizing at the same time the number of features and the error. While most approaches assess error resorting to the average of a stochastic K-fold cross-validation, comparing averages might be misleading. In this paper, we show how feature subsets with different average error might in fact be non-separable when compared using a statistical test. Following this idea, clusters of non-separable optimal feature subsets are identified. The performance in feature selection can thus be evaluated by verifying how many of these optimal feature subsets an algorithm is able to identify. We thus propose a multi-objective optimization approach to feature selection, EvoFS, with the objectives to i. minimize feature subset size, ii. minimize test error on a 10-fold cross-validation using a specific classifier, iii. maximize the analysis of variance value of the lowest-performing feature in the set. Experiments on classification datasets whose feature subsets can be exhaustively evaluated show that our approach is able to always find the best feature subsets. Further experiments on a high-dimensional classification dataset, that cannot be exhaustively analyzed, show that our approach is able to find more optimal feature subsets than state-of-the-art feature selection algorithms.

International audience; Nowadays, the use of electrolyzers to cleanly and efficiently generate hydrogen from renewable energy sources is an attractive solution. Like fuel cell systems, DC/DC converters are needed to interface the DC bus with the electrolyzer. Classic buck converters are generally used for this purpose. However, these topologies must meet several requirements from output current ripple reduction and availability in case of electrical failures point of view. Hence, an interleaved DC/DC buck converter has been chosen as a viable solution in order to meet these requirements. With the aim of improving the availability and reliability of the DC/DC converter, the development of a power switch fault detection method is mandatory. This paper proposes a novel power switch fault detection method based on a non-linear dimensionality reduction technique, implemented by means of a neural network. Experimental tests are carried out for demonstrating the performance of the proposed strategy to accurately detect the faulty leg.

International audience; This article provides an overview of the recent advances in the characterization and modeling of power-line communication (PLC) channels in transportation systems. The salient aspects of the topological and functional features of the data channels using power networks of motor vehicles, spacecraft, and aircraft are presented. This article is tutorial in nature and guides the reader through a selection of recent papers, collecting relevant results needed to assess the feasibility and strengths of PLC for this class of applications.

International audience; Carbon nanotubes (CNT) and nano-graphite (NG) are graphene-based nanomaterials which share exceptional physicochemical properties, but whose health impacts are unfortunately still not well understood. On the other hand, carbon black (CB) is a conventional and widely studied material. The comparison of these three carbon-based nanomaterials is thus of great interest to improve our understanding of their toxicity. An acid functionalization was carried out on CNT, NG and CB so that, after a thorough characterization, their impacts on RAW 264.7 macrophages could be compared for a similar surface chemistry (15 to 120 μg·mL− 1 nanomaterials, 90-min to 24-h contact). Functionalized nanomaterials triggered a weak cytotoxicity similar to the pristine nanomaterials. Acid functionalization increased the pro-inflammatory response except for CB which did not trigger any TNF-α production before or after functionalization, and seemed to strongly decrease the oxidative stress. The toxicological impact of acid functionalization appeared thus to follow a similar trend whatever the carbon-based nanomaterial. At equivalent dose expressed in surface and equivalent surface chemistry, the toxicological responses from murine macrophages to NG were higher than for CNT and CB. It seemed to correspond to the hypothesis of a platelet and fiber paradigm.

National audience; The thermodynamical study of the CaO-Al2O3-CO2-H2O system, applied to the hydration of monocalcium aluminate at ambient temperature, allows to explain both the experimental precipitation of unstable hydrates (C2AH8 and C4ACH11) and the apparent complexity of conductimetric curves, by the manifestation of supersaturation phenomena. The role of some anhydrous phases (C12A7 and CA2) present in aluminous cements, and particularly their influence on the acceleration or delay in the precipitation process, are also clearly pointed out from the observation of phase diagrams.

A sol-gel process produced microsphere of Al2O3 containing 10 wt% ZrO2 by coprecipitating Al and Zr hydroxides from the respective chlorides through the addition of NH4OH. In this paper, gel processing is examined as a function of temperature, pH, and residual Cl- content. The washed gel was peptized at 80°C with HCl and the sol concentrated to an Al 3+ ion content of 4 mol/L. The concentrated sol is used to feed a pilot plant for microsphere production. The microspheres calcined at 1200°C are dense, with an average size of 20 µm and a BET surface area of 30 m2/g.

The more favourable experimental conditions of cooling and grinding was defined for calcium aluminates obtained by solid-solid reaction with C/A ratio in the range 0.96 – 1.04. The composition of the samples was related to the apparition of different phases : “substoichiometric” solid including CA and CA 2 , with a calcium deficiency “super-stoichiometric” samples including C 12 A 7 with an excess of calcium as well as mixtures of sub and super-stoichiometric solids could be obtained. “Super-stoichiometric” samples presented ionized anionic vacancies. During hydration, “super-stoichiometric” aluminate led to the formation of a first hydrate. The hydration kinetics depended on the ratio C 12 A 7 /CA 2 of the samples.

National audience; The comminution of monocalcium aluminate depends both on the grinding process and the composition of the solid : an "Aurec" vibrating crushing mill is more acting than a planetary ball mill. The subsequent reactivity of the solid versus water is more enhanced for "super-stoichiometric" aluminate (C/A=1.04) than for "sub-stoichiometric" one. Moreover, a connection between both the "balling fitness" of the solid and its activation during grinding was pointed out, what let suppose that entities responsible of the balling phenomenon take a prominent part in the hydration kinetics and can be considered as aluminate vacancies precedently identified.

International audience; Our limited understanding of the mechanisms pertaining to the force exerted by debris flows on barriers makes it difficult to ascertain whether a design is inadequate, adequate, or over-designed. The main scientific challenge is because flow-type landslides impacting a rigid barrier is rarely captured in the field, and no systematic, physical experimental data is available to reveal the impact mechanisms. An important consideration in flow-structure interaction is that the impact dynamics can differ radically depending on the composition of the flow. Currently, no framework exists that can characterize the impact behavior for a wide range of flow compositions. This review paper examines recent works on debris-flow structure interactions and the limitations of commonly used approaches to estimate the impact load for the design of barriers. Key challenges faced in this area and outlook for further research are discussed.