HAL-Ecole des Ponts ParisTech
Not a member yet
42743 research outputs found
Sort by
The role of turbulence in the deposition of intrinsically buoyant MPs
No full textInternational audienceIntrinsically floating microplastics (MP) such as polyethene (PE) or polypropylene (PP) are among the most common MPs found in aquatic sediments. There must hence be mechanisms that cause lighter-than-water MPs to deposit despite them being buoyant. How these MPs end up in the sediment bed is only partly understood. This study explores how turbulence in the water can affect the vertical movement of buoyant MP and bring them in contact with the bed. The deposition of PE (995 kg m−3) in slow-flowing water (average flow velocities of 1.85 and 4.17 cm s−1) was measured by tracking them and analyzing their motion in an open, rectangular, glass-sided flume. Flow characteristics in terms of turbulent kinetic energy and shear velocity were measured by particle image velocimetry. Experiments were conducted at a water depth of 27 cm and at various hydraulic conditions created by adjusting inflow speeds and using different bed materials: medium gravel, fine gravel, medium sand, cohesive sediment, and glass. The results showed that the vertical velocity of the MPs in the turbulent flow regimes varied over 4 orders of magnitude from their predicted rising velocity in quiescent water (laminar flow). Turbulence mixing resulted in distribution throughout the water column with a substantial quantity consistently subject to downward vertical transport, which in turn increased the chance of the PE particles encountering the bed and potentially getting immobilized. This work provides a plausible explanation and further experimental validation for the concept of mixing induced transfer of MPs from the water surface to the sediments of shallow waters
Unsupervised discovery of Interpretable Visual Concepts
No full textInternational audienceProviding interpretability of deep-learning models to non-experts, while fundamental for a responsible real-world usage, is challenging. Attribution maps from xAI techniques, such as Integrated Gradients, are a typical example of a visualization technique containing a high level of information, but with difficult interpretation. In this paper, we propose two methods, Maximum Activation Groups Extraction (MAGE) and Multiscale Interpretable Visualization (Ms-IV), to explain the model’s decision, enhancing global interpretability. MAGE finds, for a given CNN, combinations of features which, globally, form a semantic meaning, that we call concepts. We group these similar feature patterns by clustering in “concepts”, that we visualize through Ms-IV. This last method is inspired by Occlusion and Sensitivity analysis (incorporating causality), and uses a novel metric, called Class-aware Order Correlation (CaOC), to globally evaluate the most important image regions according to the model’s decision space. We compare our approach to xAI methods such as LIME and Integrated Gradients. Experimental results evince the Ms-IV higher localization and faithfulness values. Finally, qualitative evaluation of combined MAGE and Ms-IV demonstrates humans’ ability to agree, based on the visualization, with the decision of clusters’ concepts; and, to detect, among a given set of networks, the existence of bias
Inexpensive polynomial-degree-robust equilibrated flux a posteriori estimates for isogeometric analysis
No full textInternational audienceWe consider isogeometric discretizations of the Poisson model problem, focusing on high polynomial degrees and strong hierarchical refinements. We derive a posteriori error estimates by equilibrated fluxes, i.e, vector-valued mapped piecewise polynomials lying in the H(\div) space which appropriately approximate the desired divergence constraint. Our estimates are constant-free in the leading term, locally efficient, and robust with respect to the polynomial degree. They are also robust with respect to the number of hanging nodes arising in adaptive mesh refinement employing hierarchical B-splines.Two partitions of unity are designed, one with larger supports corresponding to the mapped splines, and one with small supports corresponding to mapped piecewise multilinear finite element hat basis functions. The equilibration is only performed on the small supports, avoiding the higher computational price of equilibration on the large supports or even the solution of a global system. Thus, the derived estimates are also as inexpensive as possible.An abstract framework for such a setting is developed, whose application to a specific situation only requests a verification of a few clearly identified assumptions.Numerical experiments illustrate the theoretical developments
Articles, software, data: An Open Science ethological study
No full textInternational audienceBackground. Open Science seeks to render research outputs visible, accessible, reusable. The Open Science framework is currently evolving vigorously due, among others reasons, to the UNESCO Open Science Recommendation adopted in November 2021. In this context, it is relevant to better visualize and describe the relationships that hold among the direct protagonists of this changing landscape: research teams and their research outputs, namely: articles, software and data, as their comprehension will certainly contribute to foster better Open Science practices.Method. In this work we review and describe, through the information collected in a large number of bibliographic references, the current changing trends involving some essential, defining, characteristics and behaviors of the main components of the scientific production, namely, research teams and three kinds of research outputs they produce in many scientific areas. This comparative study is based, among others, in our recent work on the evolving concepts of research software, research data in the context of Open Science.Results. In this work we observe and document some key features in this evolving landscape such as the changing and extended roles of research team members; the need to develop a new citing and referencing culture for articles, but specially for research software and data; the rising relevance of open access (to publications, software, data) policies all over the world; the existence of some barriers and difficulties like the regulations concerning academic research close to industry, or other technological applications; the need to develop standards for the “right to be forgotten”; the need to consider the impact of Open Science costs for less favored communities, countries, institutions...Conclusions. This calls for the urgent need to observe and depict further this changing Open Science ecosystem, and to propose –as we have partially attempted in this work– new concepts to analyze this context as well as to contribute to ongoing research-on-research and to improve the implementation of Open Science practices, in order to foster better ways towards a sound, inclusive and fairer Open Science landscape.</div
Sur les ratios de la mobilité urbaine, partie 2: L'espace viaire est-il une ressource rare ? Un modèle physique et son application à cinq villes en France
No full textUrban streets are laid out as longitudinal stretches of space and their width is split into sidewalks for pedestrians and lanes for vehicle running and parking. The specific width assigned to each function determines its flowing capacity. Considering the respective spatial footprints of modal vehicles, urban planners demonstrate that pedestrians and bikes use public space more efficiently than cars. Yet, traffic engineers and economists keep to modal trip flow rates as the key indicators of the streets' social utility. The article has a twofold objective first of unifying the respective measurements of planners and economists and, second, of investigating the urban conditions under which street capacity is a scarce resource. The first objective is achieved by relating vehicle spatial footprint to vehicular spatial concentration, and vehicle time-area footprint (TAF) to vehicular flow rate. At the link level the demand-supply ratios of TAF and trip flow rate are fairly equivalent. The second objective is addressed at the city level by modeling urban conditions both on the demand and supply sides of people mobility: on the demand side, population density, trip generation rate and average trip length per travel mode and per longitudinal axiality versus, on the supply side, the lateral, "way density" of longitudinal routes and their respective widths and lanes. The model is applied to five cities spanning the range of population density in urban France, considering their respective mobility parameters and street network geometry. It comes out that people density below 4,000 p/km² have some slack street space, hence no scarcity issue, whereas densities above 6,000 p/km² combined to typical street geometry require transit services on the ground and also underground beyond 10,000 p/km²
Is broader trading welfare improving for emission trading systems?
No full textEmission trading systems are cornerstone policies to reduce carbon emissions. Although economic intuition suggests that broader allowance trading should always be welfare improving, this paper proves that view can be wrong. Under an increasingly popular type of emissions trading scheme-tradable performance standards (TPS), multiple narrow markets can decrease emissions relative to a single unified market, so that restricting trade does not always harm welfare. We show analytically that, when intensity benchmarks are heterogeneous within a sector, this result can hold even if the well-known "implicit output subsidy" does not arise. Finally, we provide evidence that this concern is not a mere theoretical possibility but can actually be of high practical relevance. Using a general equilibrium model of China's TPS for 2020-2030, we show that broader trading results in significantly higher emissions (up to 10%), and decreases welfare relative to narrower markets when the social cost of carbon exceeds $91/tCO2
The Effects of Electronic Monitoring on Offenders and their Families
No full textElectronic monitoring (EM) has emerged as a popular tool for curbing the growth of large prison populations. Evidence on the causal effects of EM on criminal recidivism is, however, limited and it is unclear how this alternative to incarceration affects the labor supply of offenders and the outcomes of their family members. We study the countrywide expansion of EM in Sweden in 1997 wherein offenders sentenced to up to three months in prison were granted the option to substitute incarceration with EM. Our difference-in-differences estimates, which compare the change in the prison inflow rate of treated offenders to that of non-treated offenders with slightly longer sentences, show that the reform significantly decreased the number of incarcerations. Our main finding is that EM not only lowers criminal recidivism but also increases labor supply. Additionally, EM improves the educational attainment and early-life earnings of the children whose parents were exposed to the reform. The primary mechanisms through which EM operates appear to involve the preservation of offenders’ ties to the labor market, by reducing the barriers to both finding a job and changing employers. Our calculations suggest that the social benefits stemming from EM are about seven times larger than the fiscal savings associated with reduced prison expenditures, implying that the welfare gains from EM could be much greater than previously acknowledged
Convex optimization for binary tree-based transport networks
No full textInternational audienceOptimizing transport networks is a well-known class of problems that have been extensively studied, with application in many domains. Here we are interested in a generalization of the Steiner problem,which entails finding a graph minimizing a cost function associated withconnecting a given set of points. In this paper, we concentrate on a specific formulation of this problem which is applied to the generation ofsynthetic vascular trees. More precisely, we focus on the ConstrainedConstructive Optimization (CCO) tree algorithm, which constructs avascular network iteratively, optimizing a blood transport energy efficiency. We show that the classical incremental construction method often leads to sub-optimal results, and that a better global solution can bereached
A time-space integro-differential economic model of epidemic control
No full textInternational audienceIn this paper we propose a time-space economic model to control the evolution and the spread of a disease. The underlying epidemiological model is formulated as a reaction-diffusion integro-differential partial differential equation. This specific model formulation, supported by empirical data, contains three different terms: a pure diffusion term, a linear growth term, and an integral term. These three terms capture different diffusion channels of a transmissible disease: a local diffusion effect, a temporal effect, and a global diffusion effect. The decision maker aims at deciding the optimal effort to be implemented in order to control the number of infections and, at the same time, minimize the cost of treatment. We analyze the finite horizon case in detail and we provide the closed-form expression of the optimal policy to be implemented to control the epidemic while sustaining economic growth. We also propose two different extensions: The first one considers an infinite horizon model while, the second one, is related to a multi-period framework
Assessing hierarchies by their consistent segmentations
No full textInternational audienceCurrent approaches to generic segmentation start by creating a hierarchy of nested image partitions and then specifying a segmentation from it. Our first contribution is to describe several ways, most of them new, for specifying segmentations using the hierarchy elements. Then, we consider the best hierarchy-induced segmentation specified by a limited number of hierarchy elements. We focus on a common quality measure for binary segmentations, the Jaccard index (also known as IoU). Optimizing the Jaccard index is highly non-trivial, and yet we propose an efficient approach for doing exactly that. This way we get algorithm-independent upper bounds on the quality of any segmentation created from the hierarchy. We found that the obtainable segmentation quality varies significantly depending on the way that the segments are specified by the hierarchy elements, and that representing a segmentation with only a few hierarchy elements is often possible. (Code is available)