Laboratoire Hubert Curien

In this paper, we introduce a new domain adaptation (DA) algorithm where the source and target domains are represented by subspaces described by eigenvectors. In this context, our method seeks a domain adaptation solution by learning a mapping function which aligns the source subspace with the target one. We show that the solution of the corresponding optimization problem can be obtained in a simple closed form, leading to an extremely fast algorithm. We use a theoretical result to tune the unique hyper parameter corresponding to the size of the subspaces. We run our method on various datasets and show that, despite its intrinsic simplicity, it outperforms state of the art DA methods.

Image denoising is an important problem in image processing since noise may interfere with visual or automatic interpretation. This paper presents a new approach for image denoising in the case of a known uncorrelated noise model. The proposed filter is an extension of the nonlocal means (NL means) algorithm introduced by Buades , which performs a weighted average of the values of similar pixels. Pixel similarity is defined in NL means as the Euclidean distance between patches (rectangular windows centered on each two pixels). In this paper, a more general and statistically grounded similarity criterion is proposed which depends on the noise distribution model. The denoising process is expressed as a weighted maximum likelihood estimation problem where the weights are derived in a data-driven way. These weights can be iteratively refined based on both the similarity between noisy patches and the similarity of patches extracted from the previous estimate. We show that this iterative process noticeably improves the denoising performance, especially in the case of low signal-to-noise ratio images such as synthetic aperture radar (SAR) images. Numerical experiments illustrate that the technique can be successfully applied to the classical case of additive Gaussian noise but also to cases such as multiplicative speckle noise. The proposed denoising technique seems to improve on the state of the art performance in that latter case.

Abstract The current concern for fat intake in western countries has raised the question of the individual fatty acid (FA) impact on health. This important issue has strengthened the awareness of nutritionists and food manufacturers for the control of the FA profile of food products. The aim of this review is to provide a classification of the FA profiles of 80 vegetable oil sources, according to their nutritional potential. The first part of the review focuses on lipoprotein metabolism, and on the impact of each dietary FA on blood lipid composition (LDL‐cholesterol, HDL‐cholesterol and circulating triacylglycerols). In the second part of the review, the oil sources are clustered by similar FA profiles, and the classification is discussed with regard to the individual FA action on blood lipid composition. Apart from the major vegetable seeds, the clustering highlighted some interesting nutritional oil sources containing mainly α‐linolenic acid (camelina, linseed, perilla and stock oils), or interesting amounts of the two essential FA (purslane, chia, raspberry seed, sea buckthorn seed and salicorn oils). Furthermore, this classification provides a useful tool for the formulation of the FA profile of food products.

The need for appropriate ways to measure the distance or similarity between data is ubiquitous in machine learning, pattern recognition and data mining, but handcrafting such good metrics for specific problems is generally difficult. This has led to the emergence of metric learning, which aims at automatically learning a metric from data and has attracted a lot of interest in machine learning and related fields for the past ten years. This survey paper proposes a systematic review of the metric learning literature, highlighting the pros and cons of each approach. We pay particular attention to Mahalanobis distance metric learning, a well-studied and successful framework, but additionally present a wide range of methods that have recently emerged as powerful alternatives, including nonlinear metric learning, similarity learning and local metric learning. Recent trends and extensions, such as semi-supervised metric learning, metric learning for histogram data and the derivation of generalization guarantees, are also covered. Finally, this survey addresses metric learning for structured data, in particular edit distance learning, and attempts to give an overview of the remaining challenges in metric learning for the years to come.

Abstract The so‐called ‘flat optics’ that shape the amplitude and phase of light with high spatial resolution are presently receiving considerable attention. Numerous journal publications seemingly offer hope for great promises for ultra‐flat metalenses with high efficiency, high numerical aperture, broadband operation… We temperate the expectation by referring to the current status of metalenses against their historical background, assessing the technical and scientific challenges recently solved and critically identifying those that still stand in the way. image

In this review paper, we present radiation effects on silica-based optical fibers. We first describe the mechanisms inducing microscopic and macroscopic changes under irradiation: radiation-induced attenuation, radiation-induced emission and compaction. We then discuss the influence of various parameters related to the optical fiber, to the harsh environments and to the fiber-based applications on the amplitudes and kinetics of these changes. Then, we focus on advances obtained over the last years. We summarize the main results regarding the fiber vulnerability and hardening to radiative constraints associated with several facilities such as Megajoule class lasers, ITER, LHC, nuclear power plants or with space applications. Based on the experience gained during these projects, we suggest some of the challenges that will have to be overcome in the near future to allow a deeper integration of fibers and fiber-based sensors in radiative environments.

Speckle noise is an inherent problem in coherent imaging systems such as synthetic aperture radar. It creates strong intensity fluctuations and hampers the analysis of images and the estimation of local radiometric, polarimetric, or interferometric properties. Synthetic aperture radar (SAR) processing chains thus often include a multilooking (i.e., averaging) filter for speckle reduction, at the expense of a strong resolution loss. Preservation of point-like and fine structures and textures requires to adapt locally the estimation. Nonlocal (NL)-means successfully adapt smoothing by deriving data-driven weights from the similarity between small image patches. The generalization of nonlocal approaches offers a flexible framework for resolution-preserving speckle reduction. We describe a general method, i.e., NL-SAR, that builds extended nonlocal neighborhoods for denoising amplitude, polarimetric, and/or interferometric SAR images. These neighborhoods are defined on the basis of pixel similarity as evaluated by multichannel comparison of patches. Several nonlocal estimations are performed, and the best one is locally selected to form a single restored image with good preservation of radar structures and discontinuities. The proposed method is fully automatic and handles single and multilook images, with or without interferometric or polarimetric channels. Efficient speckle reduction with very good resolution preservation is demonstrated both on numerical experiments using simulated data, airborne, and spaceborne radar images. The source code of a parallel implementation of NL-SAR is released with this paper.

Precise nanostructuration of surface and the subsequent upgrades in material properties is a strong outcome of ultra fast laser irradiations. Material characteristics can be designed on mesoscopic scales, carrying new optical properties. We demonstrate in this work, the possibility of achieving material modifications using ultra short pulses, via polarization dependent structures generation, that can generate specific color patterns. These oriented nanostructures created on the metal surface, called ripples, are typically smaller than the laser wavelength and in the range of visible spectrum. In this way, a complex colorization process of the material, involving imprinting, calibration and reading, has been performed to associate a priori defined colors. This new method based on the control of the laser-driven nanostructure orientation allows cumulating high quantity of information in a minimal surface, proposing new applications for laser marking and new types of identifying codes.

Silica-based optical fibers, fiber-based devices and optical fiber sensors are today integrated in a variety of harsh environments associated with radiation constraints. Under irradiation, the macroscopic properties of the optical fibers are modified through three main basic mechanisms: the radiation induced attenuation, the radiation induced emission and the radiation induced refractive index change. Depending on the fiber profile of use, these phenomena differently contribute to the degradation of the fiber performances and then have to be either mitigated for radiation tolerant systems or exploited to design radiation detectors and dosimeters. Considering the strong impact of radiation on key applications such as data transfer or sensing in space, fusion and fission-related facilities or high energy physics facilities, since 1970′s numerous experimental and theoretical studies have been conducted to identify the microscopic origins of these changes. The observed degradation can be explained through the generation by ionization or displacement damages of point defects in the differently doped amorphous glass (SiO2) of the fiber's core and cladding layers. Indeed, the fiber chemical composition (dopants/concentrations) and elaboration processes play an important role. Consequently, identifying the nature, the properties and the generation and bleaching mechanisms of these point defects is mandatory in order to imagine ways to control the fiber radiation behaviors. In this review paper, the responses of the main classes of silica-based optical fibers are presented: radiation tolerant pure-silica core or fluorine doped optical fibers, germanosilicate optical fibers and radiation sensitive phosphosilicate and aluminosilicate optical fibers. Our current knowledge about the nature and optical properties of the point defects related to silica and these main dopants is presented. The efficiency of the known defects to reproduce the transient and steady state radiation induced attenuation between 300 nm and 2 µm wavelength range is discussed. The main parameters, related to the fibers themselves or extrinsic - harsh environments, profile of use - affecting the concentration, growth and decay kinetics of those defects are also reviewed. Finally, the main remaining challenges are discussed, including the increasing needs for accurate and multi-physics modeling tools.

We present an experimental and numerical study of the damage and ablation thresholds at the surface of a dielectric material, e.g., fused silica, using short pulses ranging from 7 to 300 fs. The relevant numerical criteria of damage and ablation thresholds are proposed consistently with experimental observations of the laser irradiated zone. These criteria are based on lattice thermal melting and electronic cohesion temperature, respectively. The importance of the three major absorption channels (multi-photon absorption, tunnel effect, and impact ionization) is investigated as a function of pulse duration (7--300 fs). Although the relative importance of the impact ionization process increases with the pulse duration, our results show that it plays a role even at short pulse duration ($<$50 fs). For few optical cycle pulses (7 fs), it is also shown that both damage and ablation fluence thresholds tend to coincide due to the sharp increase of the free electron density. This electron-driven ablation regime is of primary interest for thermal-free laser-matter interaction and therefore for the development of high quality micromachining processes.

<para xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> We present a comprehensive study of the strain and temperature-sensing characteristics of single-mode–multimode–single-mode (SMS) structures based on the modal interference of guided modes of graded index multimode fiber (MMF) section spliced in between two single-mode fibers. A detailed theoretical study of the structures in terms of the refractive index distribution, effect of dopant and their concentrations, and the variation of core diameter has been carried out. Our study shows that for the SMS structure with a <formula formulatype="inline"><tex Notation="TeX">${\rm GeO}_{2}$</tex></formula>-doped MMF there exists a critical wavelength on either side of which the spectrum shows opposite spectral shift with a change in temperature/strain, whereas for structures with a <formula formulatype="inline"><tex Notation="TeX">${\rm P}_{2}{\rm O}_{5}$</tex></formula>-doped MMF it shows monotonic red shift with increasing temperature/strain. It has been found that the critical wavelength shifts toward higher wavelengths with decreasing “<formula formulatype="inline"> <tex Notation="TeX">$q$</tex></formula>” value/doping concentration. Using different MMFs, both the red and blue spectral shifts have been observed experimentally. It has also been found that the SMS structure has higher sensitivity toward this critical wavelength. The study should find application in designing strain-insensitive high-sensitive temperature sensors or vice versa. </para>

Interferometric synthetic aperture radar (SAR) data provide reflectivity, interferometric phase, and coherence images, which are paramount to scene interpretation or low-level processing tasks such as segmentation and 3-D reconstruction. These images are estimated in practice from a Hermitian product on local windows. These windows lead to biases and resolution losses due to the local heterogeneity caused by edges and textures. This paper proposes a nonlocal approach for the joint estimation of the reflectivity, the interferometric phase, and the coherence images from an interferometric pair of coregistered single-look complex (SLC) SAR images. Nonlocal techniques are known to efficiently reduce noise while preserving structures by performing the weighted averaging of similar pixels. Two pixels are considered similar if the surrounding image patches are “resembling.” Patch similarity is usually defined as the Euclidean distance between the vectors of graylevels. In this paper, a statistically grounded patch-similarity criterion suitable to SLC images is derived. A weighted maximum likelihood estimation of the SAR interferogram is then computed with weights derived in a data-driven way. Weights are defined from the intensity and interferometric phase and are iteratively refined based both on the similarity between noisy patches and on the similarity of patches from the previous estimate. The efficiency of this new interferogram construction technique is illustrated both qualitatively and quantitatively on synthetic and true data.

International audience

The scent of roses (Rosa x hybrida) is composed of hundreds of volatile molecules. Monoterpenes represent up to 70% percent of the scent content in some cultivars, such as the Papa Meilland rose. Monoterpene biosynthesis in plants relies on plastid-localized terpene synthases. Combining transcriptomic and genetic approaches, we show that the Nudix hydrolase RhNUDX1, localized in the cytoplasm, is part of a pathway for the biosynthesis of free monoterpene alcohols that contribute to fragrance in roses. The RhNUDX1 protein shows geranyl diphosphate diphosphohydrolase activity in vitro and supports geraniol biosynthesis in planta.

The sensitivity of grating-coupled Surface Plasmon Polaritons (SPPs) on metallic surface has been exploited to investigate the correlation between ripples formation under ultrashort laser exposure and SPPs generation conditions. Systematic examination of coupling of single ultrashort laser pulse on gratings with appropriate periods ranging from 440 nm to 800 nm has been performed. Our approach reveals that a surface plasmon is excited only for an appropriate grating period, the nickel sample exhibits fine ripples pattern, evidencing the plasmonic nature of ripples generation. We propose a systematic investigation supported by a comprehensive study on the obtained modulation of such a coupling efficiency by means of a phenomenological Drude-Lorentz model which captures possible optical properties modification under femtosecond irradiation.

The joint W3C (World Wide Web Consortium) and OGC (Open Geospatial Consortium) Spatial Data on the Web (SDW) Working Group developed a set of ontologies to describe sensors, actuators, samplers as well as their observations, actuation, and sampling activities. The ontologies have been published bot h as a W3C recommendation and as an OGC implementation standard. The set includes a lightweight core module called SOSA (Sensor, Observation, Sampler, and Actuator) available at: http://www.w3.org/ns/sosa/, and a more expressive extension module called SSN (Semantic Sensor Network) available at: http://www.w3.org/ns/ssn/. Together they describe systems of sensors and actuators, observations, the used procedures, the subjects and their properties being observed or acted upon, samples and the process of sampling, and so forth. The set of ontologies adopts a modular architecture with SOSA as a self-contained core that is extended by SSN and other modules to add expressivity and breadth. The SOSA/SSN ontologies are able to support a wide range of applications and use cases, including satellite imagery, large-scale scientific monitoring, industrial and household infrastructures, social sensing, citizen science, observation-driven ontology engineering, and the Internet of Things. In this paper we give an overview of the ontologies and discuss the rationale behind key design decisions, reporting on the differences between the new SSN ontology presented here and its predecessor [Web Semantics: Science, Services and Agents on the World Wide Web 17 (2012), 25–32] developed by the W3C Semantic Sensor Network Incubator group (the SSN-XG). We present usage examples and describe alignment modules that foster interoperability with other ontologies.

This paper presents GridNet, a new Convolutional Neural Network (CNN)\narchitecture for semantic image segmentation (full scene labelling). Classical\nneural networks are implemented as one stream from the input to the output with\nsubsampling operators applied in the stream in order to reduce the feature maps\nsize and to increase the receptive field for the final prediction. However, for\nsemantic image segmentation, where the task consists in providing a semantic\nclass to each pixel of an image, feature maps reduction is harmful because it\nleads to a resolution loss in the output prediction. To tackle this problem,\nour GridNet follows a grid pattern allowing multiple interconnected streams to\nwork at different resolutions. We show that our network generalizes many well\nknown networks such as conv-deconv, residual or U-Net networks. GridNet is\ntrained from scratch and achieves competitive results on the Cityscapes\ndataset.\n

The frictional properties of a rough contact interface are controlled by its area of real contact, the dynamical variations of which underlie our modern understanding of the ubiquitous rate-and-state friction law. In particular, the real contact area is proportional to the normal load, slowly increases at rest through aging, and drops at slip inception. Here, through direct measurements on various contacts involving elastomers or human fingertips, we show that the real contact area also decreases under shear, with reductions as large as 30[Formula: see text], starting well before macroscopic sliding. All data are captured by a single reduction law enabling excellent predictions of the static friction force. In elastomers, the area-reduction rate of individual contacts obeys a scaling law valid from micrometer-sized junctions in rough contacts to millimeter-sized smooth sphere/plane contacts. For the class of soft materials used here, our results should motivate first-order improvements of current contact mechanics models and prompt reinterpretation of the rate-and-state parameters.

Periodic self-organization of matter beyond the diffraction limit is a puzzling phenomenon, typical both for surface and bulk ultrashort laser processing. Here we compare the mechanisms of periodic nanostructure formation on the surface and in the bulk of fused silica. We show that volume nanogratings and surface nanoripples having subwavelength periodicity and oriented perpendicular to the laser polarization share the same electromagnetic origin. The nanostructure orientation is defined by the near-field local enhancement in the vicinity of the inhomogeneous scattering centers. The periodicity is attributed to the coherent superposition of the waves scattered at inhomogeneities. Numerical calculations also support the multipulse accumulation nature of nanogratings formation on the surface and inside fused silica. Laser surface processing by multiple laser pulses promotes the transition from the high spatial frequency perpendicularly oriented nanoripples to the low spatial frequency ripples, parallel or perpendicular to the laser polarization. The latter structures also share the electromagnetic origin, but are related to the incident field interference with the scattered far-field of rough non-metallic or transiently metallic surfaces. The characteristic ripple appearances are predicted by combined electromagnetic and thermo-mechanical approaches and supported by SEM images of the final surface morphology and by time-resolved pump-probe diffraction measurements.

The side-channel community recently investigated a new approach, based on deep learning, to significantly improve profiled attacks against embedded systems. Previous works have shown the benefit of using convolutional neural networks (CNN) to limit the effect of some countermeasures such as desynchronization. Compared with template attacks, deep learning techniques can deal with trace misalignment and the high dimensionality of the data. Pre-processing is no longer mandatory. However, the performance of attacks depends to a great extent on the choice of each hyperparameter used to configure a CNN architecture. Hence, we cannot perfectly harness the potential of deep neural networks without a clear understanding of the network’s inner-workings. To reduce this gap, we propose to clearly explain the role of each hyperparameters during the feature selection phase using some specific visualization techniques including Weight Visualization, Gradient Visualization and Heatmaps. By highlighting which features are retained by filters, heatmaps come in handy when a security evaluator tries to interpret and understand the efficiency of CNN. We propose a methodology for building efficient CNN architectures in terms of attack efficiency and network complexity, even in the presence of desynchronization. We evaluate our methodology using public datasets with and without desynchronization. In each case, our methodology outperforms the previous state-of-the-art CNN models while significantly reducing network complexity. Our networks are up to 25 times more efficient than previous state-of-the-art while their complexity is up to 31810 times smaller. Our results show that CNN networks do not need to be very complex to perform well in the side-channel context.