Search

We investigate a method to retrieve full-complex models (Transmission Matrix and Neural Network) of a highly multimode fiber (140 LP modes/polarization) using a straightforward machine learning approach, without the need of a reference beam. The models are first validated by the high fidelity between the predicted and the experimental images in the near field and far field output planes (Pearson correlation coefficient between 97.5% and 99.1% with our trained Transmission Matrix or Neural Network). Their accuracy was further confirmed by successful 3D beam shaping, a task achievable only with a true full complex model. As a prospect, we also demonstrate the ability of our neural network architecture to model nonlinear Kerr propagation in gradient index multimode fiber and predict the output beam shape.

Numerical modeling is essential for comprehending intricate physical phenomena in different domains. To handle complexity, sensitivity analysis, particularly screening, is crucial for identifying influential input parameters. Kernel-based methods, such as the Hilbert Schmidt Independence Criterion (HSIC), are valuable for analyzing dependencies between inputs and outputs. Moreover, due to the computational expense of such models, metamodels (or surrogate models) are often unavoidable. Implementing metamodels and HSIC requires data from the original model, which leads to the need for space-filling designs. While existing methods like Latin Hypercube Sampling (LHS) are effective for independent variables, incorporating dependence is challenging. This paper introduces a novel LHS variant, Quantization-based LHS, which leverages Voronoi vector quantization to address correlated inputs. The method ensures comprehensive coverage of stratified variables, enhancing distribution across marginals. The paper outlines expectation estimators based on Quantization-based LHS in various dependency settings, demonstrating their unbiasedness. The method is applied on several models of growing complexities, first on simple examples to illustrate the theory, then on more complex environmental hydrological models, when the dependence is known or not, and with more and more interactive processes and factors. The last application is on the digital twin of a French vineyard catchment (Beaujolais region) to design a vegetative filter strip and reduce water, sediment and pesticide transfers from the fields to the river. Quantization-based LHS is used to compute HSIC measures and independence tests, demonstrating its usefulness, especially in the context of complex models.

Advancements in deep learning-based 3D object detection necessitate the availability of large-scale datasets. However, this requirement introduces the challenge of manual annotation, which is often both burdensome and time-consuming. To tackle this issue, the literature has seen the emergence of several weakly supervised frameworks for 3D object detection which can automatically generate pseudo labels for unlabeled data. Nevertheless, these generated pseudo labels contain noise and are not as accurate as those labeled by humans. In this paper, we present the first approach that addresses the inherent ambiguities present in pseudo labels by introducing an Evidential Deep Learning (EDL) based uncertainty estimation framework. Specifically, we propose MEDL-U, an EDL framework based on MTrans, which not only generates pseudo labels but also quantifies the associated uncertainties. However, applying EDL to 3D object detection presents three primary challenges: (1) relatively lower pseudolabel quality in comparison to other autolabelers; (2) excessively high evidential uncertainty estimates; and (3) lack of clear interpretability and effective utilization of uncertainties for downstream tasks. We tackle these issues through the introduction of an uncertainty-aware IoU-based loss, an evidence-aware multi-task loss function, and the implementation of a post-processing stage for uncertainty refinement. Our experimental results demonstrate that probabilistic detectors trained using the outputs of MEDL-U surpass deterministic detectors trained using outputs from previous 3D annotators on the KITTI val set for all difficulty levels. Moreover, MEDL-U achieves state-of-the-art results on the KITTI official test set compared to existing 3D automatic annotators., Comment: Accepted at ICRA 2024. Code: https://github.com/paathelb/MEDL-U

In this paper, we aim to perform sensitivity analysis of set-valued models and, in particular, to quantify the impact of uncertain inputs on feasible sets, which are key elements in solving a robust optimization problem under constraints. While most sensitivity analysis methods deal with scalar outputs, this paper introduces a novel approach for performing sensitivity analysis with set-valued outputs. Our innovative methodology is designed for excursion sets, but is versatile enough to be applied to set-valued simulators, including those found in viability fields, or when working with maps like pollutant concentration maps or flood zone maps. We propose to use the Hilbert-Schmidt Independence Criterion (HSIC) with a kernel designed for set-valued outputs. After proposing a probabilistic framework for random sets, a first contribution is the proof that this kernel is characteristic, an essential property in a kernel-based sensitivity analysis context. To measure the contribution of each input, we then propose to use HSIC-ANOVA indices. With these indices, we can identify which inputs should be neglected (screening) and we can rank the others according to their influence (ranking). The estimation of these indices is also adapted to the set-valued outputs. Finally, we test the proposed method on three test cases of excursion sets.

There are various indications that the most primitive small bodies (P, D-type asteroids, comets) have surfaces made of intimate mixtures of opaque minerals and other components (silicates, carbonaceous compounds, etc.) in the form of sub-micrometre-sized grains, smaller than the wavelength at which they are observed, so-called hyperfine grains. Here, we investigate how the Vis-NIR-MIR spectral and V-band polarimetric properties of surfaces made of hyperfine grains are influenced by the relative abundance of such hyperfine materials, having strongly different optical indexes. Mixtures of grains of olivine and iron sulfide (or anthracite), as analogues of silicates and opaque minerals present on small bodies, were prepared at different proportions. The measurements reveal that these mixtures of hyperfine grains have spectral and polarimetric Vis-NIR properties varying in strongly nonlinear ways. When present at even a few percent, opaque components dominate the Vis-NIR spectral and polarimetric properties, and mask the silicate bands at these wavelengths. The Vis-NIR spectral slope ranges from red (positive slope), for pure opaque material, to blue (negative slope) as the proportion of silicates increases, which is reminiscent of the range of spectral slopes observed on P, D, X, C- and B-types asteroids. The spectra of the darkest mixtures in the Vis-NIR exhibit the absorption bands of Si-O in olivine around 10 m in the MIR, which is observed in emission for several small bodies. This work shows that both the contrasted optical indexes of the components, and the dispersion or aggregation (depending on their relative proportions) of their hyperfine grains, induce different light scattering regimes in the Vis-NIR and MIR, as observed for primitive small bodies. The optical separation of hyperfine grains seems to be a major parameter controlling the optical properties of these objects., Comment: accepted in Icarus

The MERTIS (MErcury Radiometer and Thermal Infrared Spectrometer) instrument onboard the ESA/JAXA BepiColombo mission will provide mid-infrared data, which will be crucial to characterize the surface mineralogy of Mercury. In order to interpret the results, we are creating a database of mid infrared spectra. As part of a study of synthetic glasses which are to serve as analog materials for the interpretation of remote sensing and modeling data, we present mid infrared data for analog materials of Mercury regolith, surface and mantle compositions. In addition, we provide data for similar analogs of Earth, Moon, Venus, and Mars rocks for a coherent picture. The analog samples have been first characterized by optical microscopy, Raman spectroscopy and EMPA. Powdered size fractions (0-25 micron, 25-63 micron, 63-125 micron, and 125 -250 micron) were studied in reflectance in the mid-infrared range from 2.5 to 18 micron (550 to 2000cm-1), additional micro FTIR analyses were also obtained. Results for the size fractions of the surface and regolith analogs for Mercury show typical features for amorphous material with Christiansen Features (CF) at 8 to 8.1 micron, Reststrahlen Bands (RB) at 9.8 to 9.9 micron, and the Transparency Feature (TF) at 12 micron. The six bulk silicate Mercury analogs have varying CF positions from 8.1 to 9 micron, with RB crystalline features of various olivines dominating in most samples. Similarly, bulk silicate analogs of the other planetary bodies show glassy features for the surface analogs with CF from 7.9 micron (Earth Continental Crust) to 8.3 micron (Lunar Mare), strong RB from 9.5 micron (Earth Continental Crust ) to 10.6 micron (Lunar Mare and Highlands). TF are usually very weak for the glassy analogs.

We have analyzed 14 impact melt glass samples, covering the compositional range from highly felsic to mafic/basaltic, as part of our effort to provide mid-infrared spectra (7-14 micron) for MERTIS (Mercury Radiometer and Thermal Infrared Spectrometer), an instrument onboard of the ESA/JAXA BepiColombo mission. Since Mercury was exposed to many impacts in its history, and impact glasses are also common on other bodies, powders of tektites (Irghizite, Libyan Desert Glass, Moldavite, Muong Nong, Thailandite) and impact glasses (from the Dellen, El'gygytgyn, Lonar, Mien, Mistastin, and Popigai impact structures) were analyzed in four size fractions of (0-25, 25-63, 93-125 and 125-250 micron) from 2.5 to 19 micron in bi-directional reflectance. The characteristic Christiansen Feature (CF) is identified between 7.3 micron (Libyan Desert Glass) and 8.2 micron (Dellen). Most samples show mid-infrared spectra typical of highly amorphous material, dominated by a strong Reststrahlen Band (RB) between 8.9 micron (Libyan Desert Glass) and 10.3 micron (Dellen). Even substantial amounts of mineral fragments hardly affect this general band shape. Comparisons of the SiO2 content representing the felsic/mafic composition of the samples with the CF shows felsic/intermediate glass and tektites forming a big group, and comparatively mafic samples a second one. An additional sign of a highly amorphous state is the lack of features at wavelengths longer than about 15 micron. The tektites and two impact glasses, Irghizite and El'gygytgyn respectively, have much weaker water features than most of the other impact glasses.

This study is part of an effort to build a mid-infrared database (7-14micron) of spectra for MERTIS (Mercury Radiometer and Thermal Infrared Spectrometer), an instrument onboard of the ESA/JAXA BepiColombo space probe to be launched to Mercury in 2017. Mercury was exposed to abundant impacts throughout its history. This study of terrestrial impactites can provide estimates of the effects of shock metamorphism on the mid-infrared spectral properties of planetary materials. In this study, we focus on the Noerdlinger Ries crater in Southern Germany, a well preserved and easily accessible impact crater with abundant suevite impactites. Suevite and melt glass bulk samples from Otting and Aumuehle, as well as red suevite from Polsingen were characterized and their reflectance spectra in mid-infrared range obtained. In addition, in-situ mid-infrared spectra were made from glasses and matrix areas in thin sections. The results show similar, but distinguishable spectra for both bulk suevite and melt glass samples, as well as in-situ measurements. Impact melt glass from Aumuehle and Otting have spectra dominated by a Reststrahlen band at 9.3-9.6 micron. Bulk melt rock from Polsingen and bulk suevite and fine-grained matrix have their strongest band between 9.4 to 9.6 micron. There are also features between 8.5 and 9 micron, and 12.5 - 12.8 micron associated with crystalline phases. There is evidence of weathering products in the fine-grained matrix, such as smectites. Mercury endured many impacts with impactors of all sizes over its history. So spectral characteristics observed for impactites formed only in a single impact like in the Ries impact event can be expected to be very common on planetary bodies exposed to many more impacts in their past. We conclude that in mid-infrared remote sensing data the surface of Mercury can be expected to be dominated by features of amorphous materials.

Enceladus, an icy moon of Saturn, possesses an internal water ocean and jets expelling ocean material into space. Cassini investigations indicated that the subsurface ocean could be a habitable environment having a complex interaction with the rocky core. Further investigation of the composition of the plume formed by the jets is necessary to fully understand the ocean, its potential habitability, and what it tells us about Enceladus' origin. Moonraker has been proposed as an ESA M-class mission designed to orbit Saturn and perform multiple flybys of Enceladus, focusing on traversals of the plume. The proposed Moonraker mission consists of an ESA-provided platform, with strong heritage from JUICE and Mars Sample Return, and carrying a suite of instruments dedicated to plume and surface analysis. The nominal Moonraker mission has a duration of 13.5 years. It includes a 23-flyby segment with 189 days allocated for the science phase, and can be expanded with additional segments if resources allow. The mission concept consists in investigating: i) the habitability conditions of present-day Enceladus and its internal ocean, ii) the mechanisms at play for the communication between the internal ocean and the surface of the South Polar Terrain, and iii) the formation conditions of the moon. Moonraker, thanks to state-of-the-art instruments representing a significant improvement over Cassini's payload, would quantify the abundance of key species in the plume, isotopic ratios, and physical parameters of the plume and the surface. Such a mission would pave the way for a possible future landed mission., Comment: Accepted for publication in The Planetary Science Journal

The explainability of deep networks is becoming a central issue in the deep learning community. It is the same for learning on graphs, a data structure present in many real world problems. In this paper, we propose a method that is more optimal, lighter, consistent and better exploits the topology of the evaluated graph than the state-of-the-art methods., Comment: in French language. XXVIIIe Colloque GRETSI - Traitement du Signal et des Images, Sep 2022, Nancy, France

Nowadays, deep prediction models, especially graph neural networks, have a majorplace in critical applications. In such context, those models need to be highlyinterpretable or being explainable by humans, and at the societal scope, this understandingmay also be feasible for humans that do not have a strong prior knowledgein models and contexts that need to be explained. In the literature, explainingis a human knowledge transfer process regarding a phenomenon between an explainerand an explainee. We propose EiX-GNN (Eigencentrality eXplainer forGraph Neural Networks) a new powerful method for explaining graph neural networksthat encodes computationally this social explainer-to-explainee dependenceunderlying in the explanation process. To handle this dependency, we introducethe notion of explainee concept assimibility which allows explainer to adapt itsexplanation to explainee background or expectation. We lead a qualitative studyto illustrate our explainee concept assimibility notion on real-world data as wellas a qualitative study that compares, according to objective metrics established inthe literature, fairness and compactness of our method with respect to performingstate-of-the-art methods. It turns out that our method achieves strong results inboth aspects.

We consider chance constrained optimization where it is sought to optimize a function while complying with constraints, both of which are affected by uncertainties. The high computational cost of realistic simulations strongly limits the number of evaluations and makes this type of problems particularly challenging. In such a context, it is common to rely on Bayesian optimization algorithms. Assuming, without loss of generality, that the uncertainty comes from some of the inputs, it becomes possible to build a Gaussian process model in the joint space of design and uncertain variables. A two-step acquisition function is then used to provide, both, promising optimization variables associated to relevant uncertain samples. Our overall contribution is to correlate the constraints in the GP model and exploit this to optimally decide, at each iteration, which constraint should be evaluated and at which point. The coupled Gaussian model of the constraints relies on an output-as-input encoding. The constraint selection idea is developed by enabling that each constraint can be evaluated for a different uncertain input, thus improving the refinement efficiency. Constraints coupling and selection are gradually implement in 3 algorithm variants which are compared to a reference Bayesian approach. The results are promising in terms of convergence speed, accuracy and stability as observed on a 2, a 4 and a 27-dimensional problems.