Search

International audience; We consider the application of machine learning techniques to gain insights into the effect of problem features on algorithm performance, and to automate the task of algorithm selection for distance-based multi- and many-objective optimisation problems. This is the most extensive benchmark study of such problems to date. The problem features can be set directly by the problem generator, and include e.g. the number of variables, objectives, local fronts, and disconnected Pareto sets. Using 945 problem configurations (leading to 28 350 instances) of varying complexity, we find that the problem features and the available optimisation budget (i) affect the considered algorithms (NSGA-II, IBEA, MOEA/D, and random search) in different ways and that (ii) it is possible to recommend a relevant algorithm based on problem features.

International audience; Recent findings demonstrated that object perception is affected by the competition between action representations. Simultaneous activation of distinct structural (“grasp-to-move”) and functional (“grasp-to-use”) action representations slows down perceptual judgements on objects. At the brain level, competition reduces motor resonance effects during manipulable object perception, reflected by an extinction of μ rhythm desynchronization. However, how this competition is solved in the absence of object-directed action remains unclear. The present study investigates the role of context in the resolution of the competition between conflicting action representations during mere object perception. To this aim, thirty-eight volunteers were instructed to perform a reachability judgment task on 3D objects presented at different distances in a virtual environment. Objects were conflictual objects associated with distinct structural and functional action representations. Verbs were used to provide a neutral or congruent action context prior or after object presentation. Neurophysiological correlates of the competition between action representation were recorded using EEG. The main result showed a release of μ rhythm desynchronization when reachable conflictual objects were presented with a congruent action context. Context influenced μ rhythm desynchronization when the action context was provided prior or after object presentation in a time-window compatible with object-context integration (around 1000 ms after the presentation of the first stimulus). These findings revealed that action context biases competition between co-activated action representations during mere object perception and demonstrated that μ rhythm desynchronization may be an index of activation but also competition between action representations in perception.

International audience; Recent evidence in adults indicates that object perceptual processing is affected by the competition between action representations. In the absence of specific motor plan, reachable objects associated with distinct structural (grasping) and functional (using) actions (e.g., calculator) elicit slower judgments than objects associated with similar actions (e.g., tennis ball). This effect is believed to reflect the cost entailed by the conflict between action representations. The present study aims to identify age-related changes in this conflict cost and investigate its underlying mechanisms. Five age groups from 8 to adulthood participated (n = 119). Participants performed perceptual judgments on different 3D objects in a virtual environment in order to assess their conflict cost (Experiment 1). Action priming effects and Simon effects were further assessed in the same participants as independent indices of the ability to activate action representations and to monitor conflict, respectively (Experiments 2 and 3). Experiment 1 demonstrated that the conflict cost is present in children as young as 8 and follows a non-linear, U-shape developmental trajectory between 8 and adulthood. Experiments 2 and 3 indicated that action priming effects showed a similar U-shape curve whereas Simon effects were stable across age groups. Action priming effects further predicted conflict costs at 10. Results suggest that the conflict cost relies on the ability to activate action representations from visual objects, which witnesses important changes during early adolescence. The role of general conflict monitoring abilities in conflict cost development requires further investigation. Findings will fuel models of action selection and embodied views of development.

International audience; In the present work, the dielectric properties of recycled liquid crystals (LCs) (non-purified, purified, and doped with diamond nanoparticles at 0.05, 0.1, and 0.2 wt%) were investigated. The studied LC mixtures were obtained from industrial recycling of end-of-life LC displays presenting mainly nematic phases. Dielectric measurements were carried out at room temperature on a frequency range from 0.1 to 106 Hz using an impedance analyzer. The amplitude of the oscillating voltage was fixed at 1 V using cells with homogeneous and homeotropic alignments. Results show that the dielectric anisotropy of all purified samples presents positive values and decreases after the addition of diamond nanoparticles to the LC mixtures. DC conductivity values were obtained by applying the universal law of dielectric response proposed by Jonscher. In addition, conductivity of the doped LC mixtures is lower than that of the undoped and non-purified LC.; Dans le présent travail, les propriétés diélectriques de cristaux liquides (CL) recyclés (non purifiés, purifiés et dopés avec des nanoparticules de diamant à 0,05, 0,1 et 0,2 % en poids) ont été étudiées. Les mélanges de LC étudiés ont été obtenus à partir du recyclage industriel d'écrans LC en fin de vie présentant principalement des phases nématiques. Les mesures diélectriques ont été effectuées à température ambiante sur une plage de fréquences allant de 0,1 à 106 Hz à l'aide d'un analyseur d'impédance. L'amplitude de la tension oscillante a été fixée à 1 V en utilisant des cellules avec des alignements homogènes et homéotropes. Les résultats montrent que l'anisotropie diélectrique de tous les échantillons purifiés présente des valeurs positives et diminue après l'ajout de nanoparticules de diamant aux mélanges LC. Les valeurs de conductivité DC ont été obtenues en appliquant la loi universelle de la réponse diélectrique proposée par Jonscher. De plus, la conductivité des mélanges LC dopés est inférieure à celle des LC non dopés et non purifiés.

International audience; This paper proposes an optimization approach for solving a classical problem in cellular automata theory: the 6-states Firing Squad Synchronization Problem (FSSP). To this purpose, we introduce an original optimization function which quantifies the quality of solutions according only to the main goal of the problem without taking into account any side information about cellular automata computations. This function is used for a dedicated Iterated Local Search algorithm which finds several hundreds of new solutions of the FSSP. Note, that up to present only one human-designed solution was known which is optimal in time. Most of the new solutions found by our algorithm have lower complexity (in terms of number of transitions rules used). An analysis of the fitness landscape for FSSP explains why, counter-intuitively, local search strategy can achieve good results for FSSP.

International audience

International audience; Contextual multi-armed bandits model decision problems, where the properties of the possible decisions are initially partially known, but may become better known as time passes. Such models have numerous applications and many algorithms have been proposed to provide approximate solutions. In this paper, we propose an algorithm for computing mul-tidimensional integration problems. Such problems are very common and can be solved using the Monte-Carlo method with the stratified sampling technique. This method consists in partitioning the integration domain then randomly sampling the partitions. Our algorithm considers the selection of the best partition to sample as a multi-armed bandit problem, which can be solved using the Upper Confidence Bound technique. We have experimented this approach for several integration problems and observed faster convergence rates.

International audience

International audience; This paper presents a new Remote Hyperspectral Imaging System (RHIS) embedded on an Unmanned Aquatic Drone (UAD) for plastic detection and identification in coastal and freshwater environments. This original system, namely the Remotely Operated Vehicle of the University of Littoral Côte d’Opale (ROV-ULCO), works in a near-field of view, where the distance between the hyperspectral camera and the water surface is about 45 cm. In this paper, the new ROV-ULCO system with all its components is firstly presented. Then, a hyperspectral image database of plastic litter acquired with this system is described. This database contains hyperspectral data cubes of different plastic types and polymers corresponding to the most-common plastic litter items found in aquatic environments. An in situ spectral analysis was conducted from this benchmark database to characterize the hyperspectral reflectance of these items in order to identify the absorption feature wavelengths for each type of plastic. Finally, the ability of our original system RHIS to automatically recognize different types of plastic litter was assessed by applying different supervised machine learning methods on a set of representative image patches of marine litter. The obtained results highlighted the plastic litter classification capability with an overall accuracy close to 90%. This paper showed that the newly presented RHIS coupled with the UAD is a promising approach to identify plastic waste in aquatic environments.

International audience; Monte-Carlo Tree Search (MCTS) is a popular technique for playing multi-player games. In this paper, we propose a new method to bias the playout policy of MCTS. The idea is to prune the decisions which seem " bad " (according to the previous iterations of the algorithm) before computing each playout. Thus, the method evaluates the estimated " good " moves more precisely. We have tested our improvement for the game of Havannah and compared it to several classic improvements. Our method outperforms the classic version of MCTS (with the RAVE improvement) and the different playout policies of MCTS that we have experimented.

International audience; We propose a solution method for online vehicle routing, which integrates a machine learning routine to improve tours’ quality. Our optimization model is based on the Bertsimas et al. (2019) re-optimization approach. Two separate routines are developed. The first one uses a neural network to produce realistic pick-up times for the customers to serve. The second one relies on Q-learning in addition to random walks for the construction of the backbone graph corresponding to the instance problem of each time step. The second routine gives improved results compared to the original approach.

International audience; To capture the real-world luminance values, High Dynamic Range (HDR) image processing has been developed. HDR images have a richer content than the widely-used Standard Dynamic Range (SDR) images, and are used in a number of situations, e.g. in film industry. As HDR displays are more and more commercially available, we need to be able to process HDR images as well as SDR ones (for example, devising denoising algorithms, inpainting or anti-aliasing). The most powerful methods to process images are now deep neural networks. However, the training of such networks calls for a lot of images, and HDR images datasets are relatively small.One way to generate HDR images is inverse tone mapping operators (iTMOs). They are algorithms that expand the dynamic range of SDR images. In this paper, we propose HDR-LFNet, a novel iTMO, and its HDR training dataset. Our method merges several existing handcrafted iTMOs, combined in a supervised neural network to produce an HDR output. Our lightweight network requires less training images than state-of-the-art methods, and has faster training phase. Besides, the quality of the generated images is similar to the state-of-the-art. We present the architecture as well as the subjective and experimental evaluations of our method.

International audience; The aim of this paper is to study surrogate-assisted algorithms for expensive multiobjective combinatorial optimization problems. Targeting pseudo-boolean domains, we provide a fine-grained analysis of an optimization framework using the Walsh basis as a core surrogate model. The considered framework uses decomposition in the objective space, and integrates three different components, namely, (i) an inner optimizer for searching promising solutions with respect to the so-constructed surrogate, (ii) a selection strategy to decide which solution is to be evaluated by the expensive objectives, and (iii) the strategy used to setup the Walsh order hyper-parameter. Based on extensive experiments using two benchmark problems, namely bi-objective NK-landscapes and unconstrained binary quadratic programming problems (UBQP), we conduct a comprehensive in-depth analysis of the combined effects of the considered components on search performance, and provide evidence on the effectiveness of the proposed search strategies. As a by-product, our work shed more light on the key challenges for designing a successful surrogate-assisted multi-objective combinatorial search process.

International audience; Over the last few years, many new algorithms have been proposed to solve combinatorial problems. In this field, Monte-Carlo Tree Search (MCTS) is a generic method which performs really well on several applications; for instance, it has been used with notable results in the game of Go. To find the most promising decision, MCTS builds a search tree where the new nodes are selected by sampling the search space randomly (i.e., by Monte-Carlo simulations). However, efficient Monte-Carlo policies are generally difficult to learn. Even if an improved Monte-Carlo policy performs adequately in some games, it can become useless or harmful in other games depending on how the algorithm takes into account the tactical and the strategic elements of the game. In this article, we address this problem by studying when and why a learned Monte-Carlo policy works. To this end, we use (1) two known Monte-Carlo policy improvements (PoolRave and Last-Good-Reply) and (2) two connection games (Hex and Havannah). We aim to understand how the benefit is related (a) to the number of random simulations and (b) to the various game rules (within them, tactical and strategic elements of the game). Our results indicate that improved Monte-Carlo policies, such as PoolRave or Last-Good-Reply, work better for games with a strong tactical element for small numbers of random simulations, whereas more general policies seem to be more suited for games with a strong strategic element for higher numbers of random simulations.

International audience; Gaze behavior of virtual characters in video games and virtual reality experiences is a key factor of realism and immersion. Indeed, gaze plays many roles when interacting with the environment; not only does it indicate what characters are looking at, but it also plays an important role in verbal and non-verbal behaviors and in making virtual characters alive. Automated computing of gaze behaviors is however a challenging problem, and to date none of the existing methods are capable of producing close-to-real results in an interactive context. We therefore propose a novel method that leverages recent advances in several distinct areas related to visual saliency, attention mechanisms, saccadic behavior modelling, and head-gaze animation techniques. Our approach articulates these advances to converge on a multi-map saliency-driven model which offers real-time realistic gaze behaviors for non-conversational characters, together with additional user-control over customizable features to compose a wide variety of results. We first evaluate the benefits of our approach through an objective evaluation that confronts our gaze simulation with ground truth data using an eye-tracking dataset specifically acquired for this purpose. We then rely on subjective evaluation to measure the level of realism of gaze animations generated by our method, in comparison with gaze animations captured from real actors. Our results show that our method generates gaze behaviors that cannot be distinguished from captured gaze animations. Overall, we believe that these results will open the way for more natural and intuitive design of realistic and coherent gaze animations for real-time applications.

International audience; Video projectors are designed to project onto flat white diffuse screens. Over the last few years, projector-based systems have been used, in virtual reality applications, to light non-specific environments such as the walls of a room. However, in these situations, the images seen by the user are affected by several radiometric disturbances, such as interreflection. Radiometric compensa tion methods have been proposed to reduce the disturbance caused by interreflection, but nothing has been proposed for evaluating the phenomenon itself and the effectiveness of compensation methods. In this paper, we propose a radiosity-based method to simulate light transfer in immersive environments, from a projector to a camera (the camera gives the image a user would see in a real room). This enables us to evaluate the disturbances resulting from interreflection. We also consider the effectiveness of interreflection compensation and study the influence of several parameters (projected image, projection onto a small or large part of the room, reflectivity of the walls). Our results show that radiometric compensation can reduce the influence of interreflection but is severely limited if we project onto a large part of the walls around the user, or if all the walls are bright.

International audience; Recently, deep learning methods that integrate image features gradually became a hot development trend in fusion of multispectral and hyperspectral remote sensing images, aka multi-sharpening. Fusion of a low spatial resolution hyperspectral image (LR-HSI datacube) with its corresponding high spatial resolution multispectral image (HR-MSI datacube) to reconstruct a high spatial resolution hyperspectral image (HR-HSI) has been a significant subject in recent years. Nevertheless, it is still difficult to achieve a high quality of spatial and spectral information fusion. In this paper, we propose a Deep Multi-Scale Learning Model (called DeepSen3) of spatial-spectral information fusion based on multi-scale inception residual convolutional neural network (CNN) for more efficient hyperspectral and multispectral image fusion from ESA remote sensing satellite missions (Sentinel-2 and Sentinel-3 images). The proposed DeepSen3 fusion network was applied to Sentinel-2 MSI (13 spectral bands with a spatial resolution ranging from 10, 20 to 60 m) and Sentinel-3 OLCI (21 spectral bands with a spatial resolution of 300 m) images. Extensive experiments demonstrate that the proposed DeepSen3 network achieves the best performance (both qualitatively and quantitatively) compared with recent state-of-the-art deep learning approaches.

National audience; In this paper, we are interested in the completion of a time series of multi-spectral images from a time series of hyper-spectral ones. To that end, we propose a new deep learning approach to that end. Our main contribution lies in the error completion task which allows to improve the completion performance. We show that our proposed method is able to produce high fidelity predictions with better quality indices than state-of-the-art methods on true images taken from the CIA / LGC database and Sentinel-2 / Sentinel-3 data.; Dans cet article, nous nous intéressons à la complétion d'une série temporelle d'images satellitaires multi-spectrales à partir d'une série temporelle d'images satellitaires hyper-spectrales. Nous proposons pour cela une nouvelle approche d'apprentissage profond. Notre contribution principale réside dans la tâche de complétion de l'erreur qui permet d'améliorer la qualité de complétion. Nous montrons que l'approche proposée est capable de fournir des prédictions haute fidélité avec de meilleurs indices de qualités que les approches de la littérature, sur des images réelles issues des bases de données CIA / LGC et Sentinel-2 / Sentinel-3.

International audience; An important step in the analysis and interpretation of video scenes for recognizing scenario is the aberration corrections introduced during the image acquisition in order to provide and correct real image data. This paper presents an approach on distortion correction based on stereoscopic self calibration from images sequences by using a multi-camera system of vision (network cameras). This approach for correcting image distortion brings an elegant and robust technique with good accuracy. Without any knowledge of shooting conditions, the camera's parameters will be estimated. For this, the image key points of interest are extracted from different overlapping views of multi-camera system by local descriptor, matching is realized between the images, and then the fundamental matrix is estimated and rectified if necessary. It is finally possible to calculate the camera's extrinsic and intrinsic parameters. These geometric information of the camera are used as parameter's models of the distortion correction

International audience; An image output by a camera is generally not a faithful representation of the real scene, because it undergoes a series of radiometric disturbances during the imaging process. This paper proposes a method for obtaining a more accurate measure of the light seen by a camera. Our method requires no specific calibration apparatus and only minimal supervision. Nevertheless, it is quite comprehensive, since it accounts for response function, exposure, vignetting, spatial non-uniformity of the sensor and colour balancing. Our method works in two steps. First, the camera is calibrated off-line, in a photoquantity sense. Then, the photoquantity of any scene can be estimated in-line. Our method is therefore geared to a wide range of computer vision applications where a camera is expected to give a measurement of the visible light. The paper starts by presenting a photoquantity model of the camera imaging process. It then describes the key steps of calibration and correction method. Finally, results are given and analyzed to evaluate the relevance of our approach.

International audience; Aesthetic quality assessment for photographs is an important research topic since it can be used for a number of applications, such as image database management or image browsing. In 2012, the aesthetic visual analysis (AVA) dataset has been proposed. It has since then been used to train the majority of computational models of aesthetics assessment. We observe that AVA is mainly composed of competitive photographs which notion of aesthetics differs from other kinds of photographs, such as professional photographs. In this paper, we evaluate whether or not recent aesthetics assessment models generalize well and perform well over professional photographs. We noticed that the different models we tested behave differently on both categories, and therefore do not generalize well. Besides, we fine-tuned one of the tested model using professional photographs and the results show that this fine-tuning is effectively improving the coverage of the methods.

International audience; The design of effective features enabling the development of automated landscape-aware techniques requires to address a number of inter-dependent issues. In this paper, we are interested in contrasting the amount of budget devoted to the computation of features with respect to: (i) the effectiveness of the features in grasping the characteristics of the landscape, and (ii) the gain in accuracy when solving an unknown problem instance by means of a feature-informed automated algorithm selection approach. We consider multi-objective combinatorial landscapes where, to the best of our knowledge, no in depth investigations have been conducted so far. We study simple cost-adjustable sampling strategies for extracting different state-of-the-art features. Based on extensive experiments, we report a comprehensive analysis on the impact of sampling on landscape feature values, and the subsequent automated algorithm selection task. In particular, we identify different global trends of feature values leading to non-trivial cost-vs-accuracy trade-off(s). Besides, we provide evidence that the sampling strategy can improve the prediction accuracy of automated algorithm selection. Interestingly, this holds independently of whether the sampling cost is taken into account or not in the overall solving budget.

International audience; Nous proposons deux méthodes de résolution pour le transport dynamique, i.e. arrivage des requêtes en temps réel, suivant une synergie entre apprentissage automatique et optimisation. La méthode d'optimisation de base est une ré-optimisation selon [Bertsimas et al., 2019]. La première amélioration utilise un réseau de neurones pour exploiter le contexte des instances à résoudre et produire des temps de ramassage de points réalistes. Tandis que la deuxième amélioration fait usage de l'apprentissage par renforcement en plus de marches aléatoire pour la construction du graphe support pour l'instance du problème correspondant à chaque pas de temps. Les benchmarks considérés dans le projet sont ceux de [Bertsimas et al., 2019] pour le transport urbain et de [Skiredj, 2021] pour le transport sanitaire.

The Catopsys system is a projector-miror-camera system which can be set in any room. It is an affordable way (not expensive, easy to install and easy to use) to perform various kinds of immersive mixed reality applications. The aim of this thesis is to study and optimize the radiometric process established through such system. First of all, radiometric perturbations of the camera are studied, then a correction method, which can be implemented in the Catopsys system, is described. This enables the system to capture, quite faithfully, the real environment and to estimate the pertubations caused by the projector or the room. The behaviour of the projector and the repercussion of using the room as a projection screen are then studied. A radiometric compensation method of the projection is proposed. This method makes the displayed image be more similar to the desired one. After a geometric study of the Catopsys system, specific rendering methods, based on ray tracing, are proposed. Finally, the features provided by the system to perform mixed reality applications, as well as their integration in the system workflow are detailled. Works done during this thesis permitted us to collaborate to the conception of the Catopsys system, to evaluate the feasibility of some features, to implement some of them and to publish several papers (1 international conference, 1 national journal, 1 national conference)., Le système Catopsys est un système projecteur-miroir-caméra à placer dans une pièce d'habitation quelconque. Il constitue un moyen abordable (coût, installation, utilisation) de réaliser toutes sortes d'applications de réalité mixte immersives. L'objectif de cette thèse est d'étudier et d'optimiser la chaîne de traitement radiométrique mise en oeuvre à travers ce système. Tout d'abord, les perturbations radiométriques de la caméra sont étudiées puis une méthode de correction, applicable au système Catopsys, est mise au point. Ceci permet, de façon relativement fidèle, d'acquérir l'environnement réel et d'estimer les perturbations issues du projecteur et de la pièce. L'influence du projecteur et de la pièce, utilisée comme écran de projection, sur l'image visible par l'utilisateur, est ensuite étudiée. Une méthode de compensation de la projection, permettant de rendre l'image visible plus proche de l'image voulue, est proposée. Après une étude de la géométrie du système Catopsys, des méthodes de simulation d'éclairage spécifiques, dérivées de la méthode de lancer de rayons, sont proposées. Enfin, les possibilités fournies par le système pour des applications de réalité mixte ainsi que leur intégration dans la chaîne de traitement sont étudiées. Les travaux réalisés dans le cadre de cette thèse ont permis de participer à la mise au point du système Catopsys, d'étudier la faisabilité de différentes fonctionnalités et d'en réaliser certaines. Ils ont également donné lieu à la publication de plusieurs papiers (une conférence internationale, une revue nationale, une conférence nationale).

National audience; Depuis quelques années des systèmes caméra-projecteur sont utilisés, dans le cadre de la réalité mixte, pour éclairer des environnements non dédiés comme les murs d'une pièce d'habitation. L'objectif de ce travail est de caractériser le potentiel d'un tel système. Pour cela, nous proposons une étude en simulation des échanges énergétiques, depuis le projecteur jusqu'à la caméra. Nous considérons également l'efficacité de la compensation. Enfin, nous analysons, d'après les résultats obtenus, l'influence de différents facteurs.

The notion of P-simple points was introduced by Bertrand to conceive parallel thinning algorithms. In `A 3D fully parallel thinning algorithm for generating medial faces' (Pattern Recogn. Lett. 16:83---87, 1995), Ma proposed an algorithm for which there are objects whose topology is not preserved. In this paper, we propose a new application of P-simple points: to automatically correct Ma's algorithm.

Industry 4.0 is a revolution in manufacturing by introducing disruptive technologies such as Internet of Things (IoT) and cloud-computing into the heart of the factory. The resulting increased automation and the improved production synergy between stocks, supply chains and customer demands, come along with the threats and attacks from the Internet. Despite extensive literature on the cybersecurity topic, many actors in manufacturing factories are just realizing the impact of cybersecurity in the preservation of their business. This paper introduces step-by-step the concepts and practical aspects of an Industry 4.0 manufacturing factory that are related to cybersecurity. Based on a subdivision of a typical factory into several generic perimeters, we present the vulnerabilities and threats regarding the network and devices usually found in each perimeter. Therefore, it is more efficient to present the recent proposals of the literature regarding cybersecurity guidelines and solutions in Industry 4.0. Instead of spreading a lot of references regarding every aspect of cybersecurity, we focused on a limited number of papers among the recent references. However, for each paper, we provide the details about the purpose of the proposal, the methodology adopted, the technical solution developed and its evaluation by the authors. These solutions range from classical cybersecurity countermeasures to innovative ones, such as those based on honeypots and digital twins. In order to deliver a review also useful to non scientists, we present our guidelines along with those of some organizations involved in cybersecurity harmonization and standardization in the world.

A local optima network (LON) encodes local optima connectivity in the fitness landscape of a combinatorial optimisation problem. Recently, LONs have been studied for their fractal dimension. Fractal dimension is a complexity index where a non-integer dimension can be assigned to a pattern. This paper investigates the fractal nature of LONs and how that nature relates to metaheuristic performance on the underlying problem. We use visual analysis, correlation analysis, and machine learning techniques to demonstrate that relationships exist and that fractal features of LONs can contribute to explaining and predicting algorithm performance. The results show that the extent of multifractality and high fractal dimensions in the LON can contribute in this way when placed in regression models with other predictors. Features are also individually correlated with search performance, and visual analysis of LONs shows insight into this relationship.

International audience; The massive development of remote sensing allowed many novel applications which bring new challenges. In particular, some applications such as marine observation require a good spatial, spectral, and temporal resolution. In order to tackle the last issue, spatio-temporal fusion of remote sensing data allows to complete a time series of multispectral images from, e.g., hyperspectral images. In this paper, we propose a new deep learning approach to that end. Our main contribution lies in the error completion task which allows to improve the completion performance. We show that our proposed method is able to produce high fidelity predictions with better quality indices than state-of-the-art methods on true images taken from the CIA / LGC database and Sentinel-2 / Sentinel-3 data.

International audience

A major issue while conceiving or parameterizing an optimization heuristic is to ensure an appropriate balance between exploitation and exploration of the search. Evolution strategies and neighborhood-based metaheuristics constitute relevant high-level frameworks, which ease the problem solving but are often complex to configure. Moreover, their effective behavior, according to the particularities of the search landscapes, remains difficult to grasp. In this paper, we deeply investigate the sampled walk search algorithm, which is a local search equivalent of the $$(1,\lambda )$$ -evolution strategy, considering that the neighborhood relation describes mutation possibilities. We specifically designed experiments to better understand the behavior of such a strategy offering a fine way to deal with the exploration versus exploitation dilemma. The main contribution is the analysis of search trajectories by evaluating and visualizing both their width (exploration) and their height (exploitation). More generally, we aim at bringing insights about the behavior of the $$(1,\lambda )$$ -ES in a discrete optimization context and within a fitness landscape perspective.

International audience; In the context of gradient-free multi-modal optimization, numerous algorithms are based on restarting evolution strategies. Such an algorithm classically performs many local searches, for finding all the global optima of the objective function. The strategy used to select the restarting points (i.e. the initial points of the local searches) is a crucial step of the method. In previous works, a strategy based on reinforcement learning has been proposed: the search space is partitioned and a multi-armed bandit algorithm is used to select an interesting region to sample. This strategy significantly improves the main optimization algorithm but is limited to small dimensional problems. In this paper, we propose an algorithm tackling this problem. The main contributions are (1) a tree-based scheme for hierarchically partition the search-space and (2) a multi-armed bandit strategy for traversing this tree. Thus, a node in the tree corresponds to a region of the search-space, and its children partition this region according to one dimension. The multi-armed bandit strategy is used to traverse the tree by selecting interesting children recursively. We have experimented our algorithm on difficult multi-modal functions, with small and large dimensions. For small dimensions, we observe performances comparable to previous state-of-the-art algorithms. For large dimensions, we observe better results as well as lower memory consumption.; Dans le cadre de l'optimisation multimodale sans gradient, de nombreux algorithmes sont basés sur des stratégies d'évolution de redémarrage. Un tel algorithme effectue classiquement de nombreuses recherches locales, pour trouver tous les optima globaux de la fonction objectif. La stratégie utilisée pour sélectionner les points de redémarrage (c'est-à-dire les points initiaux des recherches locales) est une étape cruciale de la méthode. Dans les travaux précédents, une stratégie basée sur l'apprentissage par renforcement a été proposée : l'espace de recherche est partitionné et un algorithme de bandit manchot est utilisé pour sélectionner une région intéressante à échantillonner. Cette stratégie améliore considérablement l'algorithme d'optimisation principal mais se limite à des problèmes de petite dimension. Dans cet article, nous proposons un algorithme qui s'attaque à ce problème. Les principales contributions sont (1) un méthode basé sur un arbre pour partitionner hiérarchiquement l'espace de recherche et (2) une stratégie de bandit manchot pour traverser cet arbre. Ainsi, un nœud de l'arbre correspond à une région de l'espace de recherche, et ses enfants partitionnent cette région selon une dimension. La stratégie du bandit manchot est utilisée pour traverser l'arbre en sélectionnant récursivement les enfants intéressants. Nous avons expérimenté notre algorithme sur des fonctions multimodales difficiles, avec de petites et de grandes dimensions. Pour les petites dimensions, nous observons des performances comparables aux algorithmes de pointe précédents. Pour les grandes dimensions, nous observons de meilleurs résultats ainsi qu'une moindre consommation de mémoire.

Pairwise constraints, a cheaper kind of supervision information that does not need to reveal the class labels of data points, were initially suggested to enhance the performance of clustering algorithms. Recently, researchers were interested in using them for feature selection. However, in most current methods, pairwise constraints are provided passively and generated randomly over multiple algorithmic runs by which the results are averaged. This leads to the need of a large number of constraints that might be redundant, unnecessary, and under some circumstances even inimical to the algorithm’s performance. It also masks the individual effect of each constraint set and introduces a human labor-cost burden. Therefore, in this paper, we suggest a framework for actively selecting and then propagating constraints for feature selection. For that, we benefit from the graph Laplacian that is defined on the similarity matrix. We assume that when a small perturbation of the similarity value between a data couple leads to a more well-separated cluster indicator based on the second eigenvector of the graph Laplacian, this couple is definitely expected to be a pairwise query of higher and more significant impact. Constraints propagation on the other side ensures increasing supervision information while decreasing the cost of human-labor. Finally, experimental results validated our proposal in comparison to other known feature selection methods and proved to be prominent.

The multiperiod vehicle routing problem with profit (mVRPP) is a selective vehicle routing problem where the planning horizon of each vehicle is divided into several periods. The aim of solving mVRPP is to design service itineraries so that the total amount of collected profit is maximized and the travel time limit of each period is respected. This problem arises in many real life applications, as the one encountered in cash-in-transit industry. In this paper, we present a metaheuristic approach based on the particle swarm optimization algorithm (PSO) to solve the mVRPP. Our approach incorporates an efficient optimal split procedure and dedicated local search operators proposed to guarantee high search intensification. Experiments conducted on an mVRPP benchmark show that our algorithm outperforms the state of the art metaheuristic approaches in terms of performance and robustness. Our PSO algorithm determines all the already known optimal solutions within a negligible computational time and finds $$88$$ strict improvements among the $$177$$ instances of the benchmark.

Soil moisture remote sensing on a global scale has been an active area of research over the past few decades due to its essential role in agriculture and in the prediction of some natural disasters. In this regard, GNSS-Reflectometry (GNSS-R) is proven as an efficient tool for the measurement of soil moisture content using remote sensing techniques. GNSS-R is a bi-static radar technique that uses the L-band GNSS signals as sources of opportunity to characterize Earth's surface, due to the fact that the reflected signals are often affected by the properties of the reflecting surface. In the context of this work, it is important to detect and fastly reach the area of interest (reflecting surface) for which the soil moisture content shall be monitored. A GNSS-R setup onboard a gyrocopter meets all the requirements of our application. This paper is dedicated to the study of airborne GNSS-R techniques for soil moisture monitoring using a low-altitude airborne carrier with a high rate (1ms for GPS C/A) carrier-to-noise ratio (C/N0) observations. To cope with the rapid displacement of the satellites footprints along the receiver trajectory, high rate (1000 Hz rate) C/N0 observations are processed. For this purpose, real flight experimentation has taken place on October 19, 2020 for 45 min. During the flight, the gyrocopter maintained a low-altitude of approximately 315m above the ground with an average speed of 95 km/h. Based on that, the size of the major axis of the first Fresnel zones that constitute the detected footprints ranged between 1,316m for a minimum elevation angle of 3° and 15m for a maximum elevation angle of 75°. Concerning the temporal resolution of the application, the raw data were sampled at a frequency of 25MHz and the C/N0 estimates were realized at a rate of 1000Hz.During the flight, an average of 9 GPS satellites have been detected of which 4 GPS satellite signals were extensively analyzed to observe the reflectivity corresponding to land, beach, and sea reflections. After analyzing the Delay Doppler Maps which provides an image of the scattering cross-section in terms of time and frequency and consequently tracking the corresponding signals, the 1ms C/N0 estimations were derived using the in-phase components of the signals as observations. The reflected signals are then linked to the footprints of the satellites and thus to the reflecting surfaces from which each processed signal has reflected using the GPS time, attitude, and position provided by onboard sensors and the GPS time extracted from the digitized GNSS signals. The ultimate aim of this study is to obtain reflectivity measurements from high rate C/N0 observations in order to provide a soil moisture mapping of the studied area, where we notice that the signals reflected from the beach had the best reflectivity followed by sea then land reflections.

This article is dedicated to the study of airborne GNSS-R signal processing techniques for water body detection and edge localization using a low-altitude airborne carrier with high rate reflectivity measurements. A GNSS-R setup on-board a carrier with reduced size and weight was developed for this application. We develop a radar technique for automatic GNSS signal segmentation in order to differentiate in-land water body surfaces based on the reflectivity measurements associated to different areas of reflection. Such measurements are derived from the GNSS signal amplitudes. We adapt a transitional model to characterize the changes in the measurements of the reflected GNSS signals from one area to another. We propose an on-line/off-line change detection algorithm for GNSS signal segmentation. A real flight experimentation took place in the context of this work obtaining reflections from different surfaces and landforms. We show, using the airborne GNSS measurements obtained, that the proposed radar technique detects in-land water body surfaces along the flight trajectory with high temporal (50 Hz ) and spatial resolution (order of 10 to 100 m2). We also show that we can localize the edges of the detected water body surfaces at meter accuracy.

International audience; Achieving a high-resolution approximation and hitting the Pareto optimal set with some if not all members of the population is the goal for multi- and many-objective optimization problems, and more so in real-world applications where there is also the desire to extract knowledge about the problem from this set. The task requires not only to reach the Pareto optimal set but also to be able to continue discovering new solutions, even if the population is filled with them. Particularly in many-objective problems where the population may not be able to accommodate the full Pareto optimal set. In this work, our goal is to investigate some tools to understand the behavior of algorithms once they converge and how their population size and particularities of their selection mechanism aid or hinder their ability to keep finding optimal solutions. Through the use of features that look into the population composition during the search process, we will look into the algorithm’s behavior and dynamics and extract some insights. Features are defined in terms of dominance status, membership to the Pareto optimal set, recentness of discovery, and replacement of optimal solutions. Complementing the study with features, we also look at the approximation through the accumulated number of Pareto optimal solutions found and its relationship to a common metric, the hypervolume. To generate the data for analysis, the chosen problem is MNK-landscapes with settings that make it easy to converge, enumerable for instances with 3 to 6 objectives. Studied algorithms were selected from representative multi- and many-objective optimization approaches such as Pareto dominance, relaxation of Pareto dominance, indicator-based, and decomposition.

International audience; The detection of immunoglobulin G (IgG) oligoclonal bands (OCB) in cerebrospinal fluid (CSF) by isoelectric focusing (IEF) is a valuable tool for the diagnosis of multiple sclerosis. Over the last decade, the results of our clinical research have suggested that tears are a non-invasive alternative to CSF. However, since tear samples have a lower IgG concentration than CSF, a sensitive OCB detection is therefore required. We are developing the first automatic tool for IEF analysis, with a view to speeding up the current visual inspection method, removing user variability, reducing misinterpretation, and facilitating OCB quantification and follow-up studies. The removal of band distortion is a key image enhancement step in increasing the reliability of automatic OCB detection. Here, we describe a novel, fully automatic band-straightening algorithm. The algorithm is based on a correlation directional warping function, estimated using an energy minimization procedure. The approach was optimized via an innovative coupling of a hierarchy of image resolutions to a hierarchy of transformation, in which band misalignment is corrected at successively finer scales. The algorithm’s performance was assessed in terms of the bands’ standard deviation before and after straightening, using a synthetic dataset and a set of 200 lanes of CSF, tear, serum and control samples on which experts had manually delineated the bands. The number of distorted bands was divided by almost 16 for the synthetic lanes and by 7 for the test dataset of real lanes. This method can be applied effectively to different sample types. It can realign minimal contrast bands and is robust for non-uniform deformations.

International audience

The Machine Learning(ML) based solutions in manufacturing industrial contexts often require skilled resources. More practical non-expert software solutions are then desired to enhance the usability of ML algorithms. The algorithm selection and configuration is one of the most difficult tasks for users like manufacturing specialists. The identification of the most appropriate algorithm in an automatic manner is among the major research challenges to achieve optimal performance of ML tools. In this paper, we present an auto-explained Automated Machine Learning tool for Big Industrial Data(AMLBID) to better cope with the prominent challenges posed by the evolution of Big Industrial Data. It is a meta-learning based decision support system for the automated selection and tuning of implied hyperparameters for ML algorithms. Moreover, the framework is equipped with an explainer module that makes the outcomes transparent and interpretable for well-performing ML systems.

International audience; Estimating the features to be extracted from an image for classification tasks are sometimes difficult, especially if images are related to a particular kind of noise. The aim of this paper is to propose a neural network architecture named Guided-Generative Network (GGN) to extract refined information that allows to correctly quantify the noise present in a sliding window of images. GNN tends to find the desired features to address such a problem in order to emit a detection criterion of this noise. The proposed GGN is applied on photorealistic images which are rendered by Monte-Carlo methods by evaluating a large number of samples per pixel. An insufficient number of samples per pixel tends to result in residual noise which is very noticeable to humans. This noise can be reduced by increasing the number of samples, as proven by Monte-Carlo theory, but this involves considerable computational time. Finding the right number of samples needed for human observers to perceive no noise is still an open problem. The results obtained show that GGN can correctly solve the problem without prior knowledge of the noise while being competitive with existing methods.

National audience

International audience; Abstract Distinguishing sequences are widely used in finite state machine-based conformance testing to solve the state identification problem. In this paper, we address the scalability issue encountered while deriving distinguishing sequences from complete observable nondeterministic finite state machines by introducing a massively parallel MapReduce version of the well-known Exact Algorithm. To the best of our knowledge, this is the first study to tackle this task using the MapReduce approach. First, we give a concise overview of the well-known Exact Algorithm for deriving distinguishing sequences from nondeterministic finite state machines. Second, we propose a parallel algorithm for this problem using the MapReduce approach and analyze its communication cost using Afrati et al. model. Furthermore, we conduct a variety of intensive and comparative experiments on a wide range of finite state machine classes to demonstrate that our proposed solution is efficient and scalable.

This paper proposes a novel legged locomotion principle for tracking the quadruped-robot base spatial motion which can generate automatically the trotting and walking gaits without flight phases. The principle is based on a new legged locomotion generator using the robot centroidal properties without a massless-leg assumption. Our quadruped robot is newly designed with symmetric legged dynamics properties using parallelogram mechanisms, and the corresponding generalized dynamics model is built. The centroidal momentum and dynamics models are established by combining the centroidal properties of the base and each leg, in which the virtual joint between the base and each leg CoM is used instead of the real joints. The stance legs follow the base motion with respect to contact constraints. The swing legs compensate for the delay influences (to centroidal motion) by the stance legs. By combining the whole-body kinematics model and the kinematics constraints relating to the contact constraints and the parallelogram mechanism, the adaptive swing-leg motion is achieved in the running process without pre-defined trajectories. When two swing legs land at different timings, the walking gait is applied. We also propose a new state machine for switching the locomotion gaits between walking and trotting, which also serves the torque controller which determines the dimension of the legged control states and the contact Jacobian. We apply our hierarchical torque controller to enable the robot with compliant dynamics properties. Therefore, our new locomotion principle integrates the system design, motion generation and whole-body torque control together, so that this one common framework is able to handle several locomotion gaits. To verify the usefulness and validity of our new locomotion principle based legged motion generator, we run several simulations: walking, trotting, turning, recovering from big lateral push force acting on the base.

International audience

International audience

International audience; The miniaturization of hyperspectral cameras has opened a new path to capture spectral information. One such camera, called the hybrid linescan camera, requires accurate control of its movement. Contrary to classical linescan cameras, where one line is available for every band in one shot, the latter asks for multiple shots to fill a line with multiple bands. Unfortunately, the reconstruction is corrupted by a parallax effect, which affects each band differently. In this article, we propose a two-step procedure, which first reconstructs an approximate datacube in two different ways, and second, performs a corrective warping on each band based on a multiple homography framework. The second step combines different stitching methods to perform this reconstruction. A complete synthetic and experimental comparison is performed by using geometric indicators of reference points. It appears throughout the course of our experimentation that misalignment is significantly reduced but remains non-negligible at the potato leaf scale.

The aim of this paper is to push a step towards the development of a surrogate-assisted methodology for expensive optimization problems that have both a combinatorial and a multiobjective nature. We target pseudo-boolean multiobjective functions, and we provide a comprehensive study on the design of a modular framework integrating three main configurable components. The proposed framework is based on the Walsh basis as a surrogate, and on a decomposition-based evolutionary paradigm for maintaining the solution set. The three considered components are: (i) the inner optimizer used for handling the soconstructed Walsh surrogate, (ii) the selection strategy allowing to decide which solution is to be evaluated by the expensive objectives, and (iii) the strategy used to setup the Walsh order hyper-parameter. Based on a thorough empirical analysis relying on two benchmark problems, namely bi-objective NK-landscapes and UBQP problems, we show the effectiveness of the proposed framework with respect to the available budget in terms of calls to the evaluation function. More importantly, our empirical findings shed more lights on the combined effects of the investigated components on search performance, thus providing a better understanding of the key challenges for designing a successful surrogate-assisted multiobjective combinatorial search process.