Search

Learning with non-modular losses is an important problem when sets of predictions are made simultaneously. The main tools for constructing convex surrogate loss functions for set prediction are margin rescaling and slack rescaling. In this work, we show that these strategies lead to tight convex surrogates iff the underlying loss function is increasing in the number of incorrect predictions. However, gradient or cutting-plane computation for these functions is NP-hard for non-supermodular loss functions. We propose instead a novel surrogate loss function for submodular losses, the Lovasz hinge, which leads to $\mathcal {O}(p\; \log p)$ O ( p log p ) complexity with $\mathcal {O}(p)$ O ( p ) oracle accesses to the loss function to compute a gradient or cutting-plane. We prove that the Lovasz hinge is convex and yields an extension. As a result, we have developed the first tractable convex surrogates in the literature for submodular losses. We demonstrate the utility of this novel convex surrogate through several set prediction tasks, including on the PASCAL VOC and Microsoft COCO datasets.

International audience; We consider a stochastic perturbation of the phase field alpha-Navier-Stokes model with vesicle-fluid interaction. It consists in a system of nonlinear evolution partial differential equations modeling the fluid-structure interaction associated to the dynamics of an elastic vesicle immersed in a moving incompressible viscous fluid. This system of equations couples a phase-field equation-for the interface between the fluid and the vesicle-to the alpha-Navier-Stokes equation-for the viscous fluid-with an extra nonlinear interaction term, namely the bending energy. The stochastic perturbation is an additive space-time noise of trace class on each equation of the system. We prove the existence and uniqueness of solution in classical spaces of $L^2$ functions with estimates of non-linear terms and bending energy. It is based on a priori estimate about the regularity of solutions of finite dimensional systems, and tightness of the approximated solution. AMS 2000 subject classifications. 60H15, 60H30, 37L55, 35Q30, 35Q35, 76D05

The present work aims at investigating the ability of a Kinetic-Based Moment Method (KBMM) to reproduce the statistics of turbulent particle-laden flows using the Anisotropic Gaussian (AG) closure. This method is the simplest KBMM member that can account for Particle Trajectory Crossing (PTC) properly with a well-posed mathematical structure Vie et al. (2015). In order to validate this model further, we investigate here 3D turbulent flows that are more representative of the mixing processes, which occurs in realistic applications. The chosen configuration is a 3D statistically-stationary Homogeneous Isotropic Turbulence (HIT) loaded with particles in a very dilute regime. The analysis focuses on the description of the first three lowest order moments of the particulate flow: the number density, the Eulerian velocity and the internal energy. A thorough numerical study on a large range of particle inertia allows us to show that the AG closure extends the ability of the Eulerian models to correctly reproduce the particle dynamics up to a Stokes number based on the Eulerian turbulence macro-scale equal to one, but also highlights the necessity of high-order numerical schemes to reach mesh convergence, especially for the number density field.

openaire: EC/FP7/600825/EU//RECONFIG In a multirobot system, using shared symbols for objects in the environment is a prerequisite for collaboration. Sharing symbols requires that each agent has anchored a symbol with an internal, sensor level representation, as well as that these symbols match between the agents. The problem can be solved easily when the internal representations can be communicated between the agents. However, with heterogeneous embodiments the available sensors are likely to differ, making it impossible to share the internal representations directly. We propose the use of pointing gestures to align symbols between a heterogeneous group of robots. We describe a planning framework that minimizes the required effort for anchoring representations across robots. The framework allows planning for both the gesturing and observing agents in a decentralized fashion. It considers both implicit sources of failure, such as ambiguous pointing, as well as costs required by actions. Simulation experiments demonstrate that the resulting planning problem has a complex solution structure with multiple local minima. Demonstration with a heterogeneous two-robot system shows the practical viability of this approach.

International audience; Blurring occurs frequently in video sequences captured by consumer devices, as a result of various factors such as lens aberrations, defocus, relative camera-scene motion, and camera shake. When it comes to the contents of archive documents such as old films and television shows, the degradations are even more serious due to several physical phenomena happening during the sensing, transmission, recording, and storing processes. We propose in this paper a versatile formulation of blind video deconvolution problems that seeks to estimate both the sharp unknown video sequence and the underlying blur kernel from an observed video. This inverse problem is severely ill-posed, and an appropriate solution can be obtained by modeling it as a nonconvex minimization problem. We provide a novel iterative algorithm to solve it, grounded on the use of recent advances in convex and nonconvex optimization techniques, and having the ability of including numerous well-known regularization strategies.

A solution is proposed to a longstanding open problem in kinetic theory, namely, given any set of realizable velocity moments up to order 2n, a closure for the moment of order 2n+1 is constructed for which the moment system found from the free-transport term in the one-dimensional (1-D) kinetic equation is globally hyperbolic and in conservative form. In prior work, the hyperbolic quadrature method of moments (HyQMOM) was introduced to close this moment system up to fourth order (n $\le$ 2). Here, HyQMOM is reformulated and extended to arbitrary even-order moments. The HyQMOM closure is defined based on the properties of the monic orthogonal polynomials Qn that are uniquely defined by the velocity moments up to order 2n -- 1. Thus, HyQMOM is strictly a moment closure and does not rely on the reconstruction of a velocity distribution function with the same moments. On the boundary of moment space, n double roots of the characteristic polynomial P2n+1 are the roots of Qn, while in the interior, P 2n+1 and Qn share n roots. The remaining n + 1 roots of P2n+1 bound and separate the roots of Qn. An efficient algorithm, based on the Chebyshev algorithm, for computing the moment of order 2n + 1 from the moments up to order 2n is developed. The analytical solution to a 1-D Riemann problem is used to demonstrate convergence of the HyQMOM closure with increasing n.

We consider a stochastic perturbation of the $\alpha$-Navier-Stokes model. The stochastic perturbation is an additive space-time noise of trace class. Under a natural condition about the trace of operator $Q$ in front of the noise, we prove the existence and uniqueness of strong solution, continuous in time in classical spaces of $L^{2}$ functions with estimates of non-linear terms. It is based on a priori estimate of solutions of finite-dimensional systems, and tightness of the approximated solution. Moreover, by studying the derivative of the solution with respect to the initial data, we can prove exponential moment of the approximated solutions, which is enough to obtain Strong Feller property and irreducibility of the transition semigroup. This leads naturally to the existence and uniqueness of an invariant measure.

International audience; A hyperbolic two-fluid model for gas-particle flow derived using the Boltzmann-Enskog kinetic theory is generalized to include added mass. In place of the virtual-mass force, to guarantee indifference to an accelerating frame of reference, the added mass is included in the mass, momentum and energy balances for the particle phase, augmented to include the portion of the particle wake moving with the particle velocity. The resulting compressible two-fluid model contains seven balance equations (mass, momentum, and energy for each phase, plus added mass) and employs a stiffened-gas model for the equation of state for the fluid. Using Sturm's theorem, the model is shown to be globally hyperbolic for arbitrary ratios of the material densities Z = ρ f /ρ p. An eight-equation extension to include the pseudo-turbulent kinetic energy (PTKE) in the fluid phase is also proposed; however, PTKE has no effect on hyperbolicity. In addition to the added mass, the key physics needed to ensure hyperbolicity for arbitrary Z is a fluid-mediated contribution to the particle-phase pressure tensor that is taken to be proportional to the volume fraction of the added mass. A numerical solver for hyperbolic equations is developed for the 1-D model, and numerical examples are employed to illustrate the behaviour of solutions to Riemann problems for different material-density ratios. The relation between the proposed two-fluid model and prior work on effective-field models is discussed, as well as possible extensions to include viscous stresses and the formulation of the model in the limit of an incompressible continuous phase.

International audience; This paper examines the Root solution of the Skorohod embedding problem given full marginals on some compact time interval. Our results are obtained by limiting arguments based on finitely-many marginals Root solution of Cox, Oblój, and Touzi. Our main result provides a characterization of the corresponding potential function by means of a convenient parabolic PDE.

International audience; The conditional quadrature method of moments (CQMOM) was introduced by Yuan and Fox [J. Comput. Phys. 230 (22), 8216–8246 (2011)] to reconstruct a velocity distribution function (VDF) from a finite set of its integer moments. The reconstructed VDF takes the form of a sum of weighted Dirac delta functions in velocity phase space, and provides a closure for the spatial flux term in the corresponding kinetic equation. The CQMOM closure for the flux leads to a weakly hyperbolic system of moment equations. In subsequent work [Chalons et al., Proceed. CTR Sum. Prog. 2010, 347–358 (2010)], the Dirac delta functions were replaced by Gaussian distributions, which make the moment system hyperbolic but at the added cost of dealing with continuous distributions. Here, a hyperbolic version of CQMOM is proposed that uses weighted Dirac delta functions. While the moment set employed for multi-Gaussian and conditional HyQMOM (CHyQMOM) are equivalent, the latter is able to access all of moment space whereas the former cannot (e.g. arbitrary values of the fourth-order velocity moment in 1-D phase space with two nodes). By making use of the properties of CHyQMOM in 2-D phase space, it is possible to control a symmetrical subset of the optimal moments [Fox, Indust. & Engng. Chem. Res. 48 (21), 9686–9696 (2009)]. Furthermore, the moment sets for 2-D problems are smaller for CHyQMOM than in the original CQMOM thanks to a judicious choice of the velocity abscissas in phase space.

We propose here to investigate the impact of small-scale effects on the bulk evolution of a two-phase flow system. More precisely, we choose to examine the sole influence of a small-scale (with respect to the bulk velocity) off-equilibrium velocity on the system. In order to narrow our analysis and avoid complex well-posedness issues, we choose to examine a simple barotropic 5-equation two-phase flow model that accounts for an equilibrium common bulk velocity and a small-scale off-equilibrium velocity. A full derivation of the model is presented: it is based on a variational principle which allows us to insert the two-scale kinematics into the model by considering two different kinetic energies. Additional entropy dissipation requirements allow us to add dissipative structures to the model. This system is neutral with respect to the topology of the flow structure and is equipped with parameters that can be connected to relaxation processes. When considering instantaneous relaxations, we obtain two limit systems of the literature that are used for the simulation of separated-phase flows. In this sense we obtain a hierarchy of models. We show that the parent 5-equation model is also compatible with the description of a bubbly fluid that allows small-scale vibrations for the disperse phase. This identification is verified and discussed through comparisons with experimental measurements of sound dispersion (Silberman, J. Acoust. Soc. Am., vol. 29, 1957, pp. 925–933; Leroyet al., J. Acoust. Soc. Am., vol. 123, 2008, pp. 1931–1940) and with the dispersion relations of a reference model for bubbly flows by Chenget al.(Trans. ASME J. Heat Transfer, vol. 107, 1985, pp. 402–408). The present work is a first contribution to a larger effort that aims at unifying models that can describe both separated and disperse two-phase flows, coupling small-scale modelling with large-scale resolution.

International audience; Multi-state physics systems (MSPS) modeling framework incorporates multi-state models that describes the systems degradation/maintenance process through transitions among discrete states, and physics-based models that describe the degradation process within the states by using physics knowledge and equations. In previous works, piecewise-deterministic Markov process (PDMP) has been adopted to treat the system dynamics and the degradation dependence in MSPS. For reliability assessment, Monte Carlo simulation and finite-volume method are two widely used numerical approaches to solve PDMP. In the present work, a comparative study considering different evaluation criteria of the two approaches is conducted on two representative case studies. We provide clear guidelines for the selection of the two approaches.

In this paper, a methodology is investigated for signal recovery in the presence of non-Gaussian noise. In contrast with regularized minimization approaches often adopted in the literature, in our algorithm the regularization parameter is reliably estimated from the observations. As the posterior density of the unknown parameters is analytically intractable, the estimation problem is derived in a variational Bayesian framework where the goal is to provide a good approximation to the posterior distribution in order to compute posterior mean estimates. Moreover, a majorization technique is employed to circumvent the difficulties raised by the intricate forms of the non-Gaussian likelihood and of the prior density. We demonstrate the potential of the proposed approach through comparisons with state-of-the-art techniques that are specifically tailored to signal recovery in the presence of mixed Poisson-Gaussian noise. Results show that the proposed approach is efficient and achieves performance comparable with other methods where the regularization parameter is manually tuned from the ground truth.

Stochastic approximation techniques play an important role in solving many problems encountered in machine learning or adaptive signal processing. In these contexts, the statistics of the data are often unknown a priori or their direct computation is too intensive, and they have thus to be estimated online from the observed signals. For batch optimization of an objective function being the sum of a data fidelity term and a penalization (e.g., a sparsity promoting function), Majorize-Minimize (MM) methods have recently attracted much interest since they are fast, highly flexible, and effective in ensuring convergence. The goal of this paper is to show how these methods can be successfully extended to the case when the data fidelity term corresponds to a least squares criterion and the cost function is replaced by a sequence of stochastic approximations of it. In this context, we propose an online version of an MM subspace algorithm and we study its convergence by using suitable probabilistic tools. Simulation results illustrate the good practical performance of the proposed algorithm associated with a memory gradient subspace, when applied to both nonadaptive and adaptive filter identification problems.

Reaction-diffusion systems involving a large number of unknowns and a wide spectrum of scales in space and time model various complex phenomena across different disciplines such as combustion science, plasma physics, or biomedical engineering. The numerical solution of these strongly multi-scale systems of partial differential equations entails specific challenges due to the potentially large stiffness stemming from the broad range of temporal scales in the nonlinear source term or from the presence of steep spatial gradients at the localized reaction fronts, thus hindering high-fidelity solutions at reasonable computational cost. A new generation of techniques featuring adaptation in space and time as well as error control has been introduced recently, yielding accurate solutions to such complex models while considering the entire spectrum of scales. Based on operator splitting, finite volume adaptive multiresolution, and high order time integrators with specific stability properties for each operator, these methods yield a high computational efficiency for stiff reaction-diffusion problems and were specifically designed to carry out large multidimensional simulations on standard architectures such as powerful workstations. While demonstrating the potential of such techniques the data structures of the original implementation, based on trees of pointers, provided limited opportunities for computational efficiency optimizations and posed challenges in terms of parallel programming and load balancing. The present contribution proposes a new implementation of the whole set of numerical methods including Radau5 and Rock4, relying on fully different data structures together with the use of a specific library, TBB, for shared-memory, task-based parallelism with work stealing. The performance of our implementation is assessed in a series of test-cases of increasing difficulty in two and three dimensions on multi-core and many-core architectures, demonstrating high scalability.

International audience; In this paper, we recall the result about the strong convergence rate of the Ninomiya-Victoir scheme and the properties of the multilevel Monte Carlo estimators involving this scheme that we introduced and studied in [2]. We are also interested in the error introduced by discretizing the ordinary differential equations involved in the Ninomiya-Victoir scheme. We prove that this error converges with strong order 2 when an explicit Runge-Kutta method with order 4 (resp. 2) is used for the ODEs corresponding to the Brownian (resp. Stratonovich drift) vector fields. We thus relax the order 5 needed in [11] for the Brownian ODEs to obtain the same order of strong convergence. Moreover, the properties of our multilevel Monte-Carlo estimators are preserved when these Runge-Kutta methods are used.; Dans cet article, nous commençons par rappeler le résultat de convergence de l’erreur forte du schéma de Ninomiya-Victoir et les propriétés des estimateurs Monte-Carlo multipas utilisant ce schéma que nous avons introduits et étudiés dans [2]. Nous nous intéressons également à l’erreur introduite en discrétisant les Équations Différentielles Ordinaires qui figurent dans le schéma de Ninomiya-Victoir lorsque leur solution n’est pas explicite. Nous montrons que l’ordre de convergence forte de cette erreur est 2 lorsque les ÉDOs correspondant aux champs de vecteurs browniens (resp. à la dérive dans l’écriture Stratonovich de l’équation différentielle stochastique) sont discrétisées avec une méthode Runge-Kutta explicite d’ordre 4 (resp. 2). L’utilisation de ces méthodes de Runge-Kutta préserve les propriétés de nos estimateurs Monte-Carlo multipas.

Integrated Deterministic and Probabilistic Safety Analysis (IDPSA) of dynamic systems calls for the development of efficient methods for accidental scenarios generation. The necessary consideration of failure events timing and sequencing along the scenarios requires the number of scenarios to be generated to increase with respect to conventional PSA. Consequently, their postprocessing for retrieving safety relevant information regarding the system behavior is challenged because of the large amount of generated scenarios that makes the computational cost for scenario postprocessing enormous and the retrieved information difficult to interpret. In the context of IDPSA, the interpretation consists in the classification of the generated scenarios as safe, failed, Near Misses (NMs), and Prime Implicants (PIs). To address this issue, in this paper we propose the use of an ensemble of Semi-Supervised Self-Organizing Maps (SSSOMs) whose outcomes are combined by a locally weighted aggregation according to two strategies: a locally weighted aggregation and a decision tree based aggregation. In the former, we resort to the Local Fusion (LF) principle for accounting the classification reliability of the different SSSOM classifiers, whereas in the latter we build a classification scheme to select the appropriate classifier (or ensemble of classifiers), for the type of scenario to be classified. The two strategies are applied for the postprocessing of the accidental scenarios of a dynamic U-Tube Steam Generator (UTSG).

International audience; High-intensity focused ultrasound (HIFU) enables the non-invasive thermal ablation of tumors. However, numerical simulations of the treatment remain complex and difficult to validate in clinically relevant situations. In this context, needle hydrophone measurements of the acoustic field downstream of seven rabbit tissue layers comprising skin, subcutaneous fat and muscle were performed in different geometrical configurations. Increasing curvature and thickness of the sample were found to decrease the focusing of the beam: typically, a curvature of 0.05 mm−1 decreased the maximum pressure by 45% and doubled the focal area. A numerical model based on k-Wave Toolbox was found to be in very good agreement with the reported measurements. It was used to extrapolate the effect of the superficial tissues on peak positive and peak negative pressure at focus, which affects both cavitation and target heating. The shape of the interface was found to have a strong influence on the values, and it is therefore an important parameter to monitor or to control in the clinical practice. This also highlights the importance of modeling realistic configurations when designing treatment procedures.

High fidelity modeling and simulation of moderately dense sprays at relatively low cost is still a major challenge for many applications. For that purpose, we introduce a new multi-fluid model based on a two-size moment formalism in sections, which are size intervals of discretization. It is derived from a Boltzmann type equation taking into account drag, evaporation and coalescence, which are representative of the complex terms that arise in multi-physics environments. The closure of the model comes from a reconstruction of the distribution. A piecewise affine reconstruction in size is thoroughly analyzed in terms of stability and accuracy, a key point for a high-fidelity and reliable description of the spray. Robust and accurate numerical methods are then developed, ensuring the realizability of the moments. The model and method are proven to describe the spray with a high accuracy in size and size-conditioned variables, resorting to a lower number of sections compared to one size moment methods. Moreover, robustness is ensured with efficient and tractable algorithms despite the numerous couplings and various algebra thanks to a tailored overall strategy. This strategy is successfully tested on a difficult 2D unsteady case, which proves the efficiency of the modeling and numerical choices.

International audience; Image registration in multimodal, multitemporal satellite imagery is one of the most important problems in remote sensing and essential for a number of other tasks such as change detection and image fusion. In this paper, inspired by the recent success of deep learning approaches we propose a novel convolutional neural network architecture that couples linear and deformable approaches for accurate alignment of remote sensing imagery. The proposed method is completely unsu-pervised, ensures smooth displacement fields and provides real time registration on a pair of images. We evaluate the performance of our method using a challenging multitempo-ral dataset of very high resolution satellite images and compare its performance with a state of the art elastic registration method based on graphical models. Both quantitative and qualitative results prove the high potentials of our method.

International audience; In this paper, we extend a previously introduced traffic cell transmission model (CTM) to take electric vehicles (EVs) and charging stations (CSs) into account. The CTM is improved in the following aspects: firstly, traffic demand of multiple origins-destinations (O-D) is considered, whereas the original CTM can not simulate the specific traffic demand of each O-D pair; secondly, CSs are integrated into CTM by developing new queueing cells and charging cells, which could also describe the specific configurations of each CS and model the queueing phenomenon in CSs; third, EVs at different states of charge are considered and, thus, the spatial-temporal charging demands in CSs can be estimated with good approximation.

International audience; In this paper we solve real-valued rough differential equations (RDEs) reflected on an irregular boundary. The solution $Y$ is constructed as the limit of a sequence $(Y^n)_{n\in\mathbb{N}}$ of solutions to RDEs with unbounded drifts $(\psi_n)_{n\in\mathbb{N}}$. The penalisation $\psi_n$ increases with $n$. Along the way, we thus also provide an existence theorem and a Doss-Sussmann representation for RDEs with a drift growing at most linearly. In addition, a speed of convergence of the sequence of penalised paths to the reflected solution is obtained. We finally use the penalisation method to prove that the law at time $t>0$ of some reflected Gaussian RDE is absolutely contiuous with respect to the Lebesgue measure.

In the framework of acoustic we systematize the derivation of nonlinear models(the Kuznetsov equation, the KZK equation and the NPE). We estimate the time for whichthe regular solutions of these models stay close of the solutions of the compressible isentropicNavier-Stokes/Euler systems (pointing out their weakest regularity) and establish similarresults between the solutions of the KZK, NPE and Westervelt equations with respectto the solutions of the Kuznetsov equation. To do so, we study the Kuznetsov equationbeginning by the Cauchy problem in the viscous case (stability, gobal well posedness ofregular solutions) and inviscid case (well posedness with optimal estimations of the maximalexistence time for regular solutions) and also in the half space with time periodic boundaryconditions or in a periodic in one direction space. We also obtain the existence and unicityof weak solutions for the strongly damped wave equation and the Westervelt equation in thelargest class of domains with irregular boundaries, along with the asymptotic convergenceof the solutions of the Westervelt equation with Robin boundary conditions on prefractalboundaries approximating a Koch mixture as fractal boundary., Dans le contexte de l’acoustique on a systématisé la dérivation de modèles nonlinéaires(l’équation de Kuznetsov, l’équation KZK et la NPE). On a estimé le temps pourlequel des solutions régulières de ces modèles restent proches des solutions des systèmes deNavier-Stokes/Euler compressibles isentropiques (en précisant leur plus faible régularité) etétabli les résultats analogues entre les solutions des équations de KZK, NPE et Westerveltpar rapport à la solution de l’équation de Kuznetsov. Pour ce faire, on a étudié l’équationde Kuznetsov en commençant par le problème de Cauchy dans les cas visqueux (stabilité,unicité et existence globale des solutions régulières) et non-visqueux (caractère bien poséavec les estimations optimales du temps d’existence maximale des solutions régulières) etégalement dans un demi espace avec des conditions au limites périodiques en temps oudans un espace périodique dans une direction. On a aussi obtenu l’existence et l’unicité dessolutions faibles pour l’équation des ondes fortement amortie et l’équation deWestervelt surla plus large classe de domaines aux bords irréguliers, ainsi que la convergence asymptotiquedes solutions de l’équation de Westervelt avec conditions de Robin sur les bords préfractalsapproximant un bord fractal de type mixture de Koch.

International audience; Elastic registration of inspiratory and expiratory MRI revealed qualitative and quantitative differences in lung deformation in study participants with systemic sclerosis compared with healthy volunteers.BackgroundCurrent imaging methods are not sensitive to changes in pulmonary function resulting from fibrosis. MRI with ultrashort echo time can be used to image the lung parenchyma and lung motion.PurposeTo evaluate elastic registration of inspiratory-to-expiratory lung MRI for the assessment of pulmonary fibrosis in study participants with systemic sclerosis (SSc).Materials and MethodsThis prospective study was performed from September 2017 to March 2018 and recruited healthy volunteers and participants with SSc and high-resolution CT (within the previous 3 months) of the chest for lung MRI. Two breath-hold, coronal, three-dimensional, ultrashort–echo-time, gradient-echo sequences of the lungs were acquired after full inspiration and expiration with a 3.0-T unit. Images were registered from inspiration to expiration by using an elastic registration algorithm. Jacobian determinants were calculated from deformation fields and represented on color maps. Similarity between areas with marked shrinkage and logarithm of Jacobian determinants less than −0.15 were compared between healthy volunteers and study participants with SSc. Receiver operating characteristic curve analysis was performed to determine the best Dice similarity coefficient threshold for diagnosis of fibrosis.ResultsSixteen participants with SSc (seven with pulmonary fibrosis at high-resolution CT) and 11 healthy volunteers were evaluated. Areas of marked shrinkage during expiration with logarithm of Jacobian determinants less than −0.15 were found in the posterior lung bases of healthy volunteers and in participants with SSc without fibrosis, but not in participants with fibrosis. The sensitivity and specificity of MRI for presence of fibrosis at high-resolution CT were 86% and 75%, respectively (area under the curve, 0.81; P = .04) by using a threshold of 0.36 for Dice similarity coefficient.ConclusionElastic registration of inspiratory-to-expiratory MRI shows less lung base respiratory deformation in study participants with systemic sclerosis–related pulmonary fibrosis compared with participants without fibrosis.

Valuing Guaranteed Minimum Withdrawal Benefit (GMWB) has attracted significant attention from both the academic field and real world financial markets. As remarked by Yang and Dai, the Black and Scholes framework seems to be inappropriate for such a long maturity products. Also Chen Vetzal and Forsyth in showed that the price of these products is very sensitive to interest rate and volatility parameters. We propose here to use a stochastic volatility model (Heston model) and a Black Scholes model with stochastic interest rate (Hull White model). For this purpose we present four numerical methods for pricing GMWB variables annuities: a hybrid tree-finite difference method and a Hybrid Monte Carlo method, an ADI finite difference scheme, and a Standard Monte Carlo method. These methods are used to determine the no-arbitrage fee for the most popular versions of the GMWB contract, and to calculate the Greeks used in hedging. Both constant withdrawal, optimal surrender and optimal withdrawal strategies are considered. Numerical results are presented which demonstrate the sensitivity of the no-arbitrage fee to economic, contractual and longevity assumptions., Comment: arXiv admin note: substantial text overlap with arXiv:1509.02686

This work focuses on the local asymptotic mixed normality (LAMN) property from high frequency observations, of a continuous time process solution of a stochastic differential equation driven by a truncated α-stable process with index α ∈ (0, 2). The process is observed on the fixed time interval [0,1] and the parameters appear in both the drift coefficient and scale coefficient. This extends the results of Clément and Gloter [Stoch. Process. Appl. 125 (2015) 2316–2352] where the index α ∈ (1, 2) and the parameter appears only in the drift coefficient. We compute the asymptotic Fisher information and find that the rate in the LAMN property depends on the behavior of the Lévy measure near zero. The proof relies on the small time asymptotic behavior of the transition density of the process obtained in Clément et al. [Preprint HAL-01410989v2 (2017)].

International audience; A series of coupled direct numerical simulations of decaying Homogeneous Isotropic Turbulence are performed on a 128 3 perodic box, with a size monodisperse population of particles at different Stokes numbers St, mass loading φ and particle number densities n 0 (mean number of particles per unit volume). Indeed, mechanisms for turbulence enhancement or suppression depend on these three combined parameters. Like previous investigators, crossover and modulations of the fluid energy spectra were observed consistently with the change in Stokes number and mass loading. Additionally, we also investigate the impact of the particle number density, which was not properly and independently characterized. This parameter plays a key role in the statistical convergence of the disperse phase and is therefore of primary importance for modeling purposes. DNS results show a clear impact of this parameter, promoting the energy at small scales while reducing the energy at large scales. Our investigation focuses on the identification of the energy transfer mechanisms, to highlight the differences between the influence of a homogeneous disperse phase (very high particle number density where statistical convergence is obtained) and heterogeneous cases (low particle number density). In particular, different regimes have been identified and described in terms of particle number density.

International audience; Various models of fatigue crack growth in different scenarios have been proposed in the literature. Here, in this paper, we propose a general prognostic framework for tracking crack evolution in equipment undergoing fatigue and predicting the Remaining Useful Life (RUL). The main contribution of this work is to integrate Particle Filtering (PF) and a new ensemble model which combines diverse physical degradation models with respect to their accuracy performance in previous time steps, in order to maximize the overall prediction capability. To validate the effectiveness of the proposed framework, a case study concerning multiple fatigue crack growth degradations is extensively investigated.

Recent challenges from severe weather events call for an adaptation in power systems planning methods, to ensure the sustainability of the resulting plans and the resilience of the power system against natural threats. Whereas studies have shown the importance of considering operational flexibility as an integral part of the design of power systems with a high share of renewable energy sources, very few works have incorporated the resilience against severe weather threats directly into the system design phase. We propose a comprehensive flexibility and resilience modeling and optimization framework for power systems planning considering extreme weather scenarios. A set of piece-wise linear models are developed to calculate the impact of extreme heat waves and drought events on the performance of the power generation units and on the system load. We propose a method to integrate the extreme climate impact within the power system planning model and investigate the results obtained compared to conventional planning methods under real future climate projections.

With the increasing penetration of electric vehicles in the transportation system, more and more interactions appear between the transportation and power systems. This requires considering the impact of disturbances in the electrified transportation system onto the stability of the power system. This paper addresses this issue by proposing a method based on a cell transmission model (CTM) of the electrified transportation system and an alternative current (AC) model of the power system. Specifically, CTM is used to simulate the dynamic realtime traffic under congestion disturbances and evaluate the charging demands in the areas of the electrified transportation system. The charging demands are input to the AC model of the power system to calculate the fluctuations in power flow distributions. The proposed method can simulate the dynamic interactions between the electrified transportation system and the power system, and quantitatively measure the impacts of traffic disturbances on the stability of the power system. A numerical example is used to illustrate the proposed method.

International audience; The reliability of steam generators in nuclear power plants has always been a challenging issue. Various diagnostic models have been proposed in the literature. However, no work has been reported on the development of a robust prediction model for forecasting the future health state of steam generators. In this paper, we propose an ARIMA-based prognostic approach for tracking the degradation evolution in a steam generator and further predicting its Remaining Useful Life (RUL) before breakdown. A case study concerning real degradation datasets from different steam generators is extensively investigated to validate the performance of the proposed model.

International audience; In this paper, an optimization model is developed for optimal design of the safety barriers in a nuclear power plant. By applying the developed model, safety barriers of different natures, i.e., the prevention, mitigation, emergency and recovery measures, can be optimized jointly for business continuity. A hierarchical numerical optimization method based on golden search and genetic algorithm is developed to obtain the optimal solutions. A numerical case study regarding the allocation of resources among the safety barriers prevention, mitigation and emergency in a nuclear power plant is worked out to maximize business continuity against the threat of steam generator tube ruptures.

This paper exposes a novel exploratory formalism, which end goal is the numerical simulation of the dynamics of a cloud of particles weakly or strongly coupled with a turbulent fluid. Giventhe large panel of expertise of the list of authors, the content of this paper scans a wide range of connexnotions, from the physics of turbulence to the rigorous definition of stochastic processes. Our approachis to develop reduced-order models for the dynamics of both carrying and carried phases which remainconsistant within this formalism, and to set up a numerical process to validate these models. Thenovelties of this paper lie in the gathering of a large panel of mathematical and physical definitionsand results within a common framework and an agreed vocabulary (sections 1 and 2), and in somepreliminary results and achievements within this context, section 3. While the first three sections havebeen simplified to the context of a gas field providing that the disperse phase only retrieves energythrough drag, the fourth section opens this study to the more complex situation when the dispersephase interacts with the continuous phase as well, in an energy conservative manner. This will allowus to expose the perspectives of the project and to conclude., ESAIM: Proceedings and Surveys, EDP Sciences, In press

Liquefied natural gas (LNG) is used as fuel in various kinds of vessels, e.g., passenger ship, ferry, cargo vessel and platform supply vessel (PSV). It is an eco-friendly bunker fuel with many advantages, like decreasing the emissions of SOX and particulate materials (PM) and meeting the international maritime organization (IMO) MARPOL Annex VI requirements on NOX emissions, and economic benefits compared to heavy fuel oil (HFO). However, the leakage of LNG-fuel is a threat for the safety of LNG-fueled vessels, due to its inflammable and explosive characteristics. This paper illustrates a framework for the quantitative risk assessment of LNG-fueled vessels with respect to potential leakage. For illustration purposes, reference is made to a typical LNG-fueled ship, as a representative case. Event tree analysis (ETA) and computational fluid dynamics (CFD) simulation are integrated for the investigation of the hazard, the analysis of the consequences, and the quantification the risk of the LNG leakage. The results of the study are used to provide risk control options (RCOs), in terms of optimal risk mitigation for LNG-fueled vessels.

Link to publishers version: 10.1016/j.ress.2016.03.004 We consider the assessment of the availability of oil and gas processing facilities operating under Arctic conditions. The novelty of the work lies in modelling the time-dependent effects of environmental conditions on the components failure and repair rates. This is done by introducing weather-dependent multiplicative factors, which can be estimated by expert judgements given the scarce data available from Arctic offshore operations. System availability is assessed considering the equivalent age of the components to account for the impacts of harsh operating conditions on component life history and maintenance duration. The application of the model by direct Monte Carlo simulation is illustrated on an oil processing train operating in Arctic offshore. A scheduled preventive maintenance task is considered to cope with the potential reductions in system availability under harsh operating conditions

International audience; Sensitivity Analysis (SA) is performed to gain fundamental insights on a system behavior that is usually reproduced by a model and to identify the most relevant input variables whose variations affect the system model functional response. For the reliability analysis of passive safety systems of Nuclear Power Plants (NPPs), models are Best Estimate (BE) Thermal Hydraulic (TH) codes, that predict the system functional response in normal and accidental conditions and, in this paper, an ensemble of three alternative invariant SA methods is innovatively set up for a SA on the TH code input variables. The ensemble aggregates the input variables raking orders provided by Pearson correlation ratio, Delta method and Beta method. The capability of the ensemble is shown on a BE-TH code of the Passive Containment Cooling System (PCCS) of an Advanced Pressurized water reactor AP1000, during a Loss Of Coolant Accident (LOCA), whose output probability density function (pdf) is approximated by a Finite Mixture Model (FMM), on the basis of a limited number of simulations.

The impact of energy production, transformation and use on the environmental resources encourage to understand the mechanisms of resource degradation and to develop proper analyses to reduce the impact of the energy systems on the environment. At the technical level, most attempts for reducing the environmental impact of energy systems focus on the improvement of process efficiency. One way toward an integrated approach is that of adopting exergy analysis for assessing efficiency and test improving design and operation solutions. The paper presents an exergy based analysis for improving efficiency and safety of energy systems, named Thermorisk analysis. The purpose of the Thermorisk analysis is to supply information to control, and eventually reduce, the risk of the systems (i.e. risk of accidents) by acting on the thermodynamic parameters and safety characteristics in the same frame. The proper combination of exergy and risk analysis allows monitoring the effects of efficiency improvement on the safety of the systems analyzed. A case study is presented, showing the potential of the analysis to identify the relation between the exergy efficiency and the risk of the system analyzed, and the contribution of inefficiencies on the safety of the process. Possible modifications in the process are indicated to improve the safety of the system.

International audience; Epileptiform discharges in interictal electroencephalography (EEG) form the mainstay of epilepsy diagnosis and localization of seizure onset. Visual analysis is rater-dependent and time consuming, especially for long-term recordings, while computerized methods can provide efficiency in reviewing long EEG recordings. This paper presents a machine learning approach for automated detection of epileptiform discharges (spikes). The proposed method first detects spike patterns by calculating similarity to a coarse shape model of a spike waveform and then refines the results by identifying subtle differences between actual spikes and false detections. Pattern classification is performed using Support Vector Machines (SVM) in a low dimensional space on which the original waveforms are embedded by Locality Preserving Projections (LPP). The automatic detection results are compared to experts' manual annotations (101 spikes) on a whole-night sleep EEG recording. The high sensitivity (97%) and the low false positive rate (0.1 min-1), calculated by intra-patient cross-validation, highlight the potential of the method for automated interictal EEG assessment.

International audience; In this work we propose a novel deep learning based pipeline for the task of brain tumor segmentation. Our pipeline consists of three primary components: (i) a preprocessing stage that exploits histogram standardization to mitigate inaccuracies in measured brain modalities, (ii) a first prediction stage that uses the V-Net deep learning architecture to output dense, per voxel class probabilities, and (iii) a prediction refinement stage that uses a Conditional Random Field (CRF) with a bilateral filtering objective for better context awareness. Additionally, we compare the V-Net architecture with a custom 3D Residual Network architecture, trained on a multi-view strategy, and our ablation experiments indicate that V-Net outperforms the 3D ResNet-18 with all bells and whistles, while fully connected CRFs as post processing, boost the performance of both networks. We report competitive results on the BraTS 2018 validation and test set.

International audience; Deep learning methods have gained increasing attention in addressing segmentation problems for medical images analysis despite the challenges inherited from the medical domain, such as limited data availability, lack of consistent textural or salient patterns, and high di-mensionality of the data. In this paper, we introduce a novel multi-network architecture that exploits domain knowledge to address those challenges. The proposed architecture consists of multiple deep neural networks that are trained after co-aligning multiple anatomies through multi-metric deformable registration. This multi-network architecture can be trained with fewer examples and leads to better performance, robustness and generalization through consensus. Comparable to human accuracy, highly promising results on the challenging task of interstitial lung disease segmentation demonstrate the potential of our approach.

Scattering networks are a class of designed Convolutional Neural Networks (CNNs) with fixed weights. We argue they can serve as generic representations for modelling images. In particular, by working in scattering space, we achieve competitive results both for supervised and unsupervised learning tasks, while making progress towards constructing more interpretable CNNs. For supervised learning, we demonstrate that the early layers of CNNs do not necessarily need to be learned, and can be replaced with a scattering network instead. Indeed, using hybrid architectures, we achieve the best results with predefined representations to-date, while being competitive with end-to-end learned CNNs. Specifically, even applying a shallow cascade of small-windowed scattering coefficients followed by 1$\times$1-convolutions results in AlexNet accuracy on the ILSVRC2012 classification task. Moreover, by combining scattering networks with deep residual networks, we achieve a single-crop top-5 error of 11.4% on ILSVRC2012. Also, we show they can yield excellent performance in the small sample regime on CIFAR-10 and STL-10 datasets, exceeding their end-to-end counterparts, through their ability to incorporate geometrical priors. For unsupervised learning, scattering coefficients can be a competitive representation that permits image recovery. We use this fact to train hybrid GANs to generate images. Finally, we empirically analyze several properties related to stability and reconstruction of images from scattering coefficients., arXiv admin note: substantial text overlap with arXiv:1703.08961

Arctic waters have historically been regarded as harsh environments owing to their extreme weather conditions and remoteness from land. The advantages of shorter sea routes and hydrocarbon energy exploitation have recently led to increased marine activities in such harsh environments. To ensure safe operation within the area, the potential risks of ship accidents, need to be systematically analyzed, assessed and managed along with the associated uncertainties. The treatment of epistemic uncertainty in the likelihoods of adverse events due to lack of knowledge and information should also be considered. This paper presents a Frank copula-based fuzzy event tree analysis approach to assess the risks of major ship accidents in Arctic waters, taking uncertainty into consideration. The quantitative approach includes four steps, namely, accident scenario modeling by an event tree model, probability and dependence analysis of the associated intermediate events, risk assessment with respect to the consequent outcome events. A major ship accident in Arctic waters - ships stuck in ice, is chosen as a case to interpret the modeling process of the approach proposed. Crews and ships owners can use such approach to defining risk control options that enable optimal risk mitigation. Maritime management may also benefit from better risk assessment.

International audience; In this paper, we study a nonconvex continuous relaxation of MAP inference in discrete Markov random fields (MRFs). We show that for arbitrary MRFs, this relaxation is tight, and a discrete stationary point of it can be easily reached by a simple block coordinate descent algorithm. In addition, we study the resolution of this relaxation using popular gradient methods, and further propose a more effective solution using a multilinear decomposition framework based on the alternating direction method of multi-pliers (ADMM). Experiments on many real-world problems demonstrate that the proposed ADMM significantly outper-forms other nonconvex relaxation based methods, and compares favorably with state of the art MRF optimization algorithms in different settings.

International audience; We present a method based on heterogeneous ensemble learning for the prediction of the Remaining Useful Life (RUL) of cutting tools (knives) used in the packaging industry. Ensemble diversity is achieved by training multiple prognostic models using different learning algorithms. The combination of the outcomes of the modelsin the ensembleis based on a weighted averaging strat-egy,which assigns weights proportional to the individual model performances on patterns of a vali-dation set. The proposed heterogeneous ensemble has been applied to real condition monitoring knife data. It has provided more accurate RUL predictions compared to those of each individual base model.

International audience; Long-range dependence (LRD) has been observed in a variety of phenomena in nature, and for several years also in the spiking activity of neurons. Often, this is interpreted as originating from a non-Markovian system. Here we show that a purely Markovian integrate-and-fire (IF) model, with a noisy slow adaptation term, can generate interspike intervals (ISIs) that appear as having LRD. However a proper analysis shows that this is not the case asymptotically. For comparison, we also consider a new model of individual IF neuron with fractional (non-Markovian) noise. The correlations of its spike trains are studied and proven to have LRD, unlike classical IF models. On the other hand, to correctly measure long-range dependence, it is usually necessary to know if the data are stationary. Thus, a methodology to evaluate stationarity of the ISIs is presented and applied to the various IF models. We explain that Markovian IF models may seem to have LRD because of non-stationarities.

The integrity of structures exhibiting flaws in Pressurized Water Reactor (PWR) has to be assessed to meet safety criteria. This paper deals with crack-propagation under cyclic thermo-mechanical loadings, as encountered in class I austenitic pipes of PWR’s. To have a conservative and reliable assessment of the crack propagation due to the in-service loading, various codes and standards use simplified method. For example, the RSE-M Code introduces a plastic correction depending on the proportion of the mechanical loading. An improvement of the current method requires additional investigations. Moreover, components loaded with transient or thermal fluctuations are not really in loadcontrolled conditions. To this end, a device called PROFATH was designed. The specimen is a pre-cracked thick-walled tube undergoing a set of thermal cycles and loaded with a static mechanical force. During the first part of the thermal cycle, a high frequency induction coil heats the external wall of the tube. Then, the heating system stops and the specimen is cooled down by running water inside the tube. Finite element calculations show that only a region half-way along the tube should be heated to ensure adequate structural effect. In the heated zone, the machining of a sharp circumferential groove ensures the propagation of a unique crack. An electro-mechanical jack controls the level of the mechanical static load. Tests have been carried out, and these tests allow having an evaluation of the pertinence of the correction proposed by the RSE-M Code for a significant plasticity.

International audience; The Regularized Nonlinear Acceleration (RNA) algorithm is an acceleration method capable of improving the rate of convergence of many optimization schemes such as gradient descend, SAGA or SVRG. Until now, its analysis is limited to convex problems, but empirical observations shows that RNA may be extended to wider settings. In this paper, we investigate further the benefits of RNA when applied to neural networks, in particular for the task of image recognition on CIFAR10 and ImageNet. With very few modifications of exiting frameworks, RNA improves slightly the optimization process of CNNs, after training.

International audience; It is widely believed that the success of deep convolutional networks is based on progressively discarding uninformative variability about the input with respect to the problem at hand. This is supported empirically by the difficulty of recovering images from their hidden representations, in most commonly used network architectures. In this paper we show via a one-to-one mapping that this loss of information is not a necessary condition to learn representations that generalize well on complicated problems, such as ImageNet. Via a cascade of homeomorphic layers, we build the i-RevNet, a network that can be fully inverted up to the final projection onto the classes, i.e. no information is discarded. Building an invertible architecture is difficult, for one, because the local inversion is ill-conditioned, we overcome this by providing an explicit inverse. An analysis of i-RevNets learned representations suggests an alternative explanation for the success of deep networks by a progressive contraction and linear separation with depth. To shed light on the nature of the model learned by the i-RevNet we reconstruct linear interpolations between natural image representations.

International audience; Thermal-Hydraulic (TH) codes are used to simulate the response of nuclear safety systems under transient and accident conditions. The outcomes of the simulations are used to verify the safety margins required for safe operation and make decisions on how to maintain them. In this work, a novel Expert System (ES) based on Regional Sensitivity Analysis (RSA) is developed to guide a system undergoing an accident scenario towards the safest conditions in the optimal number of operation. The ES proceeds by firstly identifying the (uncertain) system controllable variables (i.e., control rods position, feed-water flow rate, void fraction inside the steam generator, etc.) that most affect the system response by RSA; then, the limit-state function is calibrated on a dataset of outcomes of TH code runs and the system failure boundary (i.e., the limit surface) is defined on the set of (uncertain) TH input variables. Application of the ES is firstly shown with respect to an analytical case study that artificially simulates the response of a NPP to an accident scenario and, then, to a practical case study concerning the response of the pressurizer of a Pressurized Water Reactor (PWR).