Search

We propose risk models for a portfolio of risks, each following a compound Poisson distribution, with dependencies introduced through a family of tree-based Markov random fields with Poisson marginal distributions inspired in C\^ot\'e et al. (2024b, arXiv:2408.13649). The diversity of tree topologies allows for the construction of risk models under several dependence schemes. We study the distribution of the random vector of risks and of the aggregate claim amount of the portfolio. We perform two risk management tasks: the assessment of the global risk of the portfolio and its allocation to each component. Numerical examples illustrate the findings and the efficiency of the computation methods developed throughout. We also show that the discussed family of Markov random fields is a subfamily of the multivariate Poisson distribution constructed through common shocks., Comment: 21 pages, 8 figures

The increasingly large numbers of multiple images in cluster-scale gravitational lenses have allowed for tighter constraints on the mass distributions of these systems. Most lens models have progressed alongside this increase in image number. The general assumption is that these improvements would result in lens models converging to a common solution, suggesting that models are approaching the true mass distribution. To test whether or not this is occurring, we examine a sample of lens models of MACS J0416.1$-$2403 containing varying number of images as input. Splitting the sample into two bins (those including $<150$ and $>150$ images), we quantify the similarity of models in each bin using three comparison metrics, two of which are novel: Median Percent Difference, Frechet Distance, and Wasserstein Distance. In addition to quantifying similarity, the Frechet distance metric seems to also be an indicator of the mass sheet degeneracy. Each metric indicates that models with a greater number of input images are no more similar between one another than models with fewer input images. This suggests that lens models are neither converging nor diverging to a common solution for this system, regardless of method. With this result, we suggest that future models more carefully investigate lensing degeneracies and anomalous mass clumps (mass features significantly displaced from baryonic counterparts) to rigorously evaluate their model's validity. We also recommend further study into alternative, underutilized lens model priors (e.g. flux ratios) as an additional input constraint to image positions in hopes of breaking existing degeneracies., Comment: 10 pages, 7 figures. To be submitted to the Open Journal of Astrophysics. Comments welcome!

Parametric strong lensing studies of galaxy clusters often display misleading features: group/cluster scale dark matter components without any stellar counterpart, offsets between both components larger than what might be allowed by neither Cold Dark Matter nor self interacting Dark Matter models, or significant unexplained external shear components. I am revisiting mass models where such misleading (and interesting) features have been reported, adopting the following working hypothesis: any group or cluster scale dark matter clump introduced in the modelling should be associated with a luminous counterpart, and any well motivated and reliable prior should be considered, even when this degrades the fit. The goal is to derive a physically motivated description of the dark matter component which might be compared to theoretical expectations. I succeed doing so in galaxy clusters AS 1063, MACS J0416 and MACS J1206, finding that the shape of the inner dark matter component has a flat density profile. These findings may be useful for the interpretation within dark matter scenario, such as self-interacting dark matter. I fail in Abell 370: a three dark matter clumps mass model (each clump being associated with its stellar counterpart) is unable to reproduce the observational constraints with a precision smaller than 2.3 arcsec. In order to provide a sub arcsec precision, I need to describe the dark matter distribution using a four dark matter clumps model, as found in earlier works. Examining the total projected mass maps, I however find a good agreement between the total mass and the stellar distribution in Abell 370, both being, to first order, bimodal. I interpret the misleading features as being symptomatic of the lack of realism of a parametric description of the dark matter distribution. I encourage caution and criticism on the outputs of parametric strong lensing modelling., Comment: Proceedings of the SF2A 2024

The topology of the tree underlying a tree-structured Markov random field (MRF) is central to the understanding of its stochastic dynamics: it is, after all, what synthesizes the rich dependence relations within the MRF. In this paper, we shed light on the influence of the tree's topology, through an extensive comparison-based analysis, on the aggregate distribution of the MRF. This is done within the framework of a recently introduced family of tree-structured MRFs with the uncommon property of having fixed Poisson marginal distributions unaffected by the dependence scheme. We establish convex orderings of sums of MRFs encrypted on trees having different topologies, leading to the devising of a new poset of trees. Hasse diagrams, cataloguing trees of dimension up to 9, and methods for the comparison of higher-dimension trees are provided to offer an exhaustive investigation of the new poset. We also briefly discuss its relation to other existing posets of trees and to invariants from spectral graph theory. Such an analysis requires, beforehand, to study the joint distribution of a MRF's component and its sum, a random vector we refer to as a synecdochic pair. To assess if a component is less or more contributing than another to the sum, we employ stochastic orders to compare synecdochic pairs within a MRF. The resulting orderings are reflected through allocation-related quantities, which thus act as centrality indices., Comment: 22 pages, 12 figures

We make absolute frequency measurements of Cs Rydberg transitions, $\vert 6S_{1/2}, F=3 \rangle \rightarrow \vert nS_{1/2}~(n=23\rm{-}90)\rangle$ and $\vert nD_{3/2,5/2}~(n=21\rm{-}90)\rangle$, with an accuracy of less than $ 72\,\rm kHz$. The quantum defect parameters for the measured Rydberg series are the most precise obtained to date. The quantum defect series is terminated at $\delta_4$, showing that prior fits requiring higher order quantum defects reflect uncertainties in the observations. The precision of the measured quantum defects allow for the calculation of Rydberg electric-dipole transitions and fine-structure intervals extrapolated from high principal quantum numbers, to rival that of sophisticated many-body relativistic calculations carried out at low Rydberg principal quantum numbers. We quantitatively predict the contributions to the quantum defect parameters from core polarization and core penetration of Cs inner shell electrons. A new value for the ionization energy, consistent across the $ nS_{1/2}$ and $ nD_{3/2,5/2}$ Rydberg series, is reported at $31406.467 751 48 (14)~\rm{cm}^{-1}$., Comment: 15 pages, 7 figures, 7 tables

We report a direct determination of the cubic dielectric spectra of a realistic polar molecule, glycerol, from molecular dynamic (MD) simulations. From the liquid state to the mildly supercooled regime, we observed the emergence and growth of a hump in the cubic modulus, traditionally associated with collective effects in the dynamics. Its evolution follows that of dynamical correlations probed by the four-point susceptibility and its apparition at high temperature coincides with the onset of super-activation. We show that, for this system, the shape of cubic spectra is only weakly affected by dipolar cross-correlations. The good agreement with experimental observations, despite the difference in temperature range, demonstrates the relevance of this approach to help get an insight into the intricate effects probed by non-linear dielectric spectroscopy., Comment: 7 pages including appendix, 7 figures including appendix

Machine learning models are increasingly being deployed in real-world contexts. However, systematic studies on their transferability to specific and critical applications are underrepresented in the research literature. An important example is visual anomaly detection (VAD) for robotic power line inspection. While existing VAD methods perform well in controlled environments, real-world scenarios present diverse and unexpected anomalies that current datasets fail to capture. To address this gap, we introduce $\textit{CableInspect-AD}$, a high-quality, publicly available dataset created and annotated by domain experts from Hydro-Qu\'ebec, a Canadian public utility. This dataset includes high-resolution images with challenging real-world anomalies, covering defects with varying severity levels. To address the challenges of collecting diverse anomalous and nominal examples for setting a detection threshold, we propose an enhancement to the celebrated PatchCore algorithm. This enhancement enables its use in scenarios with limited labeled data. We also present a comprehensive evaluation protocol based on cross-validation to assess models' performances. We evaluate our $\textit{Enhanced-PatchCore}$ for few-shot and many-shot detection, and Vision-Language Models for zero-shot detection. While promising, these models struggle to detect all anomalies, highlighting the dataset's value as a challenging benchmark for the broader research community. Project page: https://mila-iqia.github.io/cableinspect-ad/., Comment: 35 pages, to appear at NeurIPS 2024

Nanotubes align molecules into one dimensional chains creating collective states through the coupling of the molecular transition dipole moments. These collective excitations have strong fluorescence, narrow bandwidth, and shifted emission/absorption energies. We study the optical properties of {\alpha}-sexithiophene chains in boron nitride nanotubes by combining fluorescence with far- and near-field absorption spectroscopy. The inner nanotube diameter determines the number of encapsulated molecular chains. A single chain of {\alpha}-sexithiophene molecules has an optical absorption and emission spectrum that is red-shifted by almost 300 meV compared to the monomer emission, which is much larger than expected from dipole-dipole coupling. The collective state splits into excitation and emission channels with a Stokes shift of 200 meV for chains with two or more files. Our study emphasises the formation of a delocalized collective state through Coulomb coupling of the transition moments that shows a remarkable tuneability in transition energy.

A new family of tree-structured Markov random fields for a vector of discrete counting random variables is introduced. According to the characteristics of the family, the marginal distributions of the Markov random fields are all Poisson with the same mean, and are untied from the strength or structure of their built-in dependence. This key feature is uncommon for Markov random fields and most convenient for applications purposes. The specific properties of this new family confer a straightforward sampling procedure and analytic expressions for the joint probability mass function and the joint probability generating function of the vector of counting random variables, thus granting computational methods that scale well to vectors of high dimension. We study the distribution of the sum of random variables constituting a Markov random field from the proposed family, analyze a random variable's individual contribution to that sum through expected allocations, and establish stochastic orderings to assess a wide understanding of their behavior., Comment: 27 pages, 10 figures

Minimizing negative impacts of Artificial Intelligent (AI) systems on human societies without human supervision requires them to be able to align with human values. However, most current work only addresses this issue from a technical point of view, e.g., improving current methods relying on reinforcement learning from human feedback, neglecting what it means and is required for alignment to occur. Here, we propose to distinguish strong and weak value alignment. Strong alignment requires cognitive abilities (either human-like or different from humans) such as understanding and reasoning about agents' intentions and their ability to causally produce desired effects. We argue that this is required for AI systems like large language models (LLMs) to be able to recognize situations presenting a risk that human values may be flouted. To illustrate this distinction, we present a series of prompts showing ChatGPT's, Gemini's and Copilot's failures to recognize some of these situations. We moreover analyze word embeddings to show that the nearest neighbors of some human values in LLMs differ from humans' semantic representations. We then propose a new thought experiment that we call "the Chinese room with a word transition dictionary", in extension of John Searle's famous proposal. We finally mention current promising research directions towards a weak alignment, which could produce statistically satisfying answers in a number of common situations, however so far without ensuring any truth value., Comment: Accepted for publication in Scientific Reports, special issue on AI aligment

We present new free-form and hybrid mass reconstructions of the galaxy cluster lens MACS J0416.1$-$2403 at $z=0.396$ using the lens inversion method GRALE. The reconstructions use 237 spectroscopically confirmed multiple images from Bergamini et. al. 2023 as the main input. Our primary model reconstructs images to a positional accuracy of 0.191", thus representing one of the most precise reconstructions of this lens to date. Our models find broad agreement with previous reconstructions, and identify two $\sim 10^{12} M_{\odot}$ light-unaffiliated substructures. We focus on two highly magnified arcs: Spock and Mothra. Our model features a unique critical curve structure around the Spock arc with 2 crossings. This structure enables sufficient magnification across this arc to potentially explain the large number of transients as microlensing events of supergiant stars. Additionally, we develop a model of the millilens substructure expected to be magnifying Mothra, which may be a binary pair of supergiants with $\mu \sim 6000$. This model accounts for flexibility in the millilens position while preserving the observed flux and minimizing image position displacements along the Mothra arc. We constrain the millilens mass and core radius to $\lesssim 10^6 M_{\odot}$ and $\lesssim 17$ pc, respectively, which would render it one of the smallest and most compact substructures constrained by lensing. If the millilens is dominated by wave dark matter, the axion mass is constrained to be $\lesssim 3.0 \times 10^{-21}$ eV. Further monitoring of this lens with JWST will uncover more transients, permitting tighter constraints on the structure surrounding these two arcs., Comment: Accepted to MNRAS

For decades, Simultaneous Ascending Auction (SAA) has been the most widely used mechanism for spectrum auctions, and it has recently gained popularity for allocating 5G licenses in many countries. Despite its relatively simple rules, SAA introduces a complex strategic game with an unknown optimal bidding strategy. Given the high stakes involved, with billions of euros sometimes on the line, developing an efficient bidding strategy is of utmost importance. In this work, we extend our previous method, a Simultaneous Move Monte-Carlo Tree Search (SM-MCTS) based algorithm named $SMS^{\alpha}$ to incomplete information framework. For this purpose, we compare three determinization approaches which allow us to rely on complete information SM-MCTS. This algorithm addresses, in incomplete framework, the four key strategic issues of SAA: the exposure problem, the own price effect, budget constraints, and the eligibility management problem. Through extensive numerical experiments on instances of realistic size with an uncertain framework, we show that $SMS^{\alpha}$ largely outperforms state-of-the-art algorithms by achieving higher expected utility while taking less risks, no matter which determinization method is chosen., Comment: 15 pages, 9 figures, 2 pseudocodes

Building on the one-to-one relationship between generalized FGM copulas and multivariate Bernoulli distributions, we prove that the class of multivariate distributions with generalized FGM copulas is a convex polytope. Therefore, we find sharp bounds in this class for many aggregate risk measures, such as value-at-risk, expected shortfall, and entropic risk measure, by enumerating their values on the extremal points of the convex polytope. This is infeasible in high dimensions. We overcome this limitation by considering the aggregation of identically distributed risks with generalized FGM copula specified by a common parameter $p$. In this case, the analogy with the geometrical structure of the class of Bernoulli distribution allows us to provide sharp analytical bounds for convex risk measures., Comment: 30 pages, 1 figure

As hydrogen is touted as a key player in the decarbonization of modern society, it is critical to enable quantitative H analysis at high spatial resolution, if possible at the atomic scale. Indeed, H has a known deleterious impact on the mechanical properties (strength, ductility, toughness) of most materials that can hinder their use as part of the infrastructure of a hydrogen-based economy. Enabling H mapping, including local hydrogen concentration analyses at specific microstructural features, is essential for understanding the multiple ways that H affect the properties of materials, including for instance embrittlement mechanisms and their synergies, but also spatial mapping and quantification of hydrogen isotopes is essential to accurately predict tritium inventory of future fusion power plants, ensuring their safe and efficient operation for example. Atom probe tomography (APT) has the intrinsic capabilities for detecting hydrogen (H), and deuterium (D), and in principle the capacity for performing quantitative mapping of H within a material's microstructure. Yet the accuracy and precision of H analysis by APT remain affected by the influence of residual hydrogen from the ultra-high vacuum chamber that can obscure the signal of H from within the material, along with a complex field evaporation behavior. The present article reports the essence of discussions at a focused workshop held at the Max-Planck Institute for Sustainable Materials in April 2024. The workshop was organized to pave the way to establishing best practices in reporting APT data for the analysis of H. We first summarize the key aspects of the intricacies of H analysis by APT and propose a path for better reporting of the relevant data to support interpretation of APT-based H analysis in materials.

We present a combined strong and weak gravitational-lensing analysis of the massive galaxy cluster MACS J1423.8+2404 ($z=0.545$, MACS J1423 hereafter), one of the most dynamically relaxed and massive cool-core clusters discovered in the MAssive Cluster Survey at $z>0.5$. We combine high-resolution imaging from the Hubble Space Telescope (HST) in the F606W, F814W, and F160W pass-bands with spectroscopic observations taken as part of the KALEIDOSCOPE survey with the Multi-Unit Spectroscopic Explorer mounted on the Very Large Telescope. Our strong lensing analysis of the mass distribution in the cluster core is constrained by four multiple-image systems (17 individual images) within redshift range $1.779

Recent advancements in deep learning have proven highly effective in medical image classification, notably within histopathology. However, noisy labels represent a critical challenge in histopathology image classification, where accurate annotations are vital for training robust deep learning models. Indeed, deep neural networks can easily overfit label noise, leading to severe degradations in model performance. While numerous public pathology foundation models have emerged recently, none have evaluated their resilience to label noise. Through thorough empirical analyses across multiple datasets, we exhibit the label noise resilience property of embeddings extracted from foundation models trained in a self-supervised contrastive manner. We demonstrate that training with such embeddings substantially enhances label noise robustness when compared to non-contrastive-based ones as well as commonly used noise-resilient methods. Our results unequivocally underline the superiority of contrastive learning in effectively mitigating the label noise challenge. Code is publicly available at https://github.com/LucasDedieu/NoiseResilientHistopathology., Comment: 16 pages

MACS J0600.1-2008 (MACS0600) is an X-ray luminous, massive galaxy cluster at $z_{\mathrm{d}}=0.43$, studied previously by the REionization LensIng Cluster Survey (RELICS) and ALMA Lensing Cluster Survey (ALCS) projects which revealed a complex, bimodal mass distribution and an intriguing high-redshift object behind it. Here, we report on the results of a combined analysis of the extended strong lensing (SL), X-ray, Sunyaev-Zeldovich (SZ), and galaxy luminosity-density properties of this system. Using new JWST and ground-based Gemini-N and Keck data, we obtain 13 new spectroscopic redshifts of multiply imaged galaxies and identify 12 new photometric multiple-image systems and candidates, including two multiply imaged $z\sim7$ objects. Taking advantage of the larger areal coverage, our analysis reveals an additional bimodal, massive SL structure which we measure spectroscopically to lie adjacent to the cluster and whose existence was implied by previous SL-modeling analyses. While based in part on photometric systems identified in ground-based imaging requiring further verification, our extended SL model suggests that the cluster may have the second-largest critical area and effective Einstein radius observed to date, $A_{\mathrm{crit}}\simeq2.16 \mathrm{arcmin}^2$ and $\theta_{\mathrm{E}}=49.7''\pm5.0''$ for a source at $z_{\mathrm{s}}=2$, enclosing a total mass of $M(<\theta_{\mathrm{E}})=(4.7\pm0.7)\times10^{14} \mathrm{M}_{\odot}$. These results are also supported by the galaxy luminosity distribution, the SZ and X-ray data. Yet another, probably related massive cluster structure, discovered in X-rays $5'$ (1.7 Mpc) further north, suggests that MACS0600 is part of an even larger filamentary structure. This discovery adds to several recent detections of massive structures around SL galaxy clusters and establishes MACS0600 as a prime target for future high-redshift surveys with JWST., Comment: Accepted for publication in MNRAS. Version updated to the accepted paper

Driving molecular assembly into micrometer-scale patterns is key for defining advanced materials of interest in various fields, including life sciences, photovoltaics, and quantum photonics. However, the driving process competes with other forces, such as Brownian motion, ripening phenomena, capillary forces, and non-specific adsorption. Here we report on a guided diffusion mechanism of luminescent dye molecules encapsulated inside boron nitride nanotubes (BNNTs). Correlative measurements between BNNT bending and molecular position along the BNNT axis reveal an efficient and long-range migration of dyes from curved to straight regions of the nanotube. This curvature activated diffusion forms clusters of bright J-aggregates in periodic patterns of well-defined spacing and length. A phenomenological model of guided molecular transport in bended BNNTs is used to describe this directed 1D diffusion inside BNNT. It is shown to accurately predict the position and morphologies of a J-aggregate as a function of nanotube length. Coupling topological stimuli to 1D molecular diffusion at the nanoscale is here presented as an interesting tool capable of reconfiguring various emissive patterns of functional molecules at the mesoscopic scale., Comment: Supplementary added page 21

There has been an increasing interest in atom probe tomography (APT) to characterise hydrated and biological materials. A major benefit of APT compared to microscopy techniques more commonly used in biology is its combination of outstanding 3D spatial resolution (~0.2 nm) and mass sensitivity. APT has already been successfully used to characterise biological materials, revealing key structural information at the atomic scale, however there are many challenges inherent to the analysis of hydrated materials. New preparation protocols, often involving sample preparation and transfer at cryogenic temperature, enable APT analysis of hydrated materials and have the potential to enable 3D atomic scale characterisation of biological materials in the near-native hydrated state. In this study, APT specimens of pure water at the tips of tungsten needles were prepared at room temperature by graphene encapsulation. A parametric study was conducted where samples were transferred at either room temperature or cryo-temperature and analysed by APT by varying parameters such as the flight path and pulsing mode. The differences between the acquisition scenarios are presented along with recommendations for future studies.

Recycled aluminum alloys are pivotal for sustainable manufacturing, offering strength, durability, and environmental advantages. However, the presence of iron (Fe) impurities poses a major challenge, undermining their properties and recyclability. Conventional manufacturing processes result in coarse Fe-rich intermetallic compounds that limit the tolerance of Fe content and negatively influence performance of advanced aluminum alloys. To address this, rapid solidification techniques like direct strip casting have been explored. In this work, a detailed study of the strip cast microstructure was conducted by scanning electron microscopy, electron backscattered diffraction and atom probe tomography. Our results reveal that alloys produced by DSC exhibit significantly refined microstructures and are free from coarse Fe-rich intermetallics, thereby retaining the majority of Fe in solid solution. These findings indicate that strip casting significantly enhances Fe-tolerance in aluminum alloys, making it an attractive process for future aluminum recycling, with implications for sustainable high-performance applications.

Unstructured data are a promising new source of information that insurance companies may use to understand their risk portfolio better and improve the customer experience. However, these novel data sources are difficult to incorporate into existing ratemaking frameworks due to the size and format of the unstructured data. In this paper, we propose a framework to use street view imagery within a generalized linear model. To do so, we use representation learning to extract an embedding vector containing useful information from the image. This embedding is dense and low-dimensional, making it appropriate to use within existing ratemaking models. We find that there is useful information included in street view imagery to predict the frequency of claims for certain types of perils. This model can be used as-is in a ratemaking framework but also opens the door to future empirical research on attempting to extract the causal effect from images that lead to increased or decreased predicted claim frequencies. Throughout, we discuss the practical difficulties (technical and social) of using this type of data for insurance pricing.

The orientational dynamics of supercooled glycerol using molecular dynamics simulations for temperatures ranging from 323 K to 253 K, is probed through correlation functions of first and second ranks of Legendre polynomials, pertaining respectively to dielectric spectroscopy (DS) and depolarized dynamic light scattering (DDLS). The self, cross, and total correlation functions are compared with relevant experimental data. The computations reveal the low sensitivity of DDLS to cross-correlations, in agreement with what is found in experimental work, and strengthen the idea of directly comparing DS and DDLS data to evaluate the effect of cross-correlations in polar liquids. The analysis of the net static cross-correlations and their spatial decomposition shows that, although cross-correlations extend over nanometric distances, their net magnitude originates, in the case of glycerol, from the first shell of neighbouring molecules. Accessing the angular dependence of the static correlation allows us to get a microscopic understanding of why the rank-1 correlation function is more sensitive to cross-correlation than its rank-2 counterpart., Comment: 9 pages, 6 figures

This article addresses the problem of Ultra Reliable Low Latency Communications (URLLC) in wireless networks, a framework with particularly stringent constraints imposed by many Internet of Things (IoT) applications from diverse sectors. We propose a novel Deep Reinforcement Learning (DRL) scheduling algorithm, named NOMA-PPO, to solve the Non-Orthogonal Multiple Access (NOMA) uplink URLLC scheduling problem involving strict deadlines. The challenge of addressing uplink URLLC requirements in NOMA systems is related to the combinatorial complexity of the action space due to the possibility to schedule multiple devices, and to the partial observability constraint that we impose to our algorithm in order to meet the IoT communication constraints and be scalable. Our approach involves 1) formulating the NOMA-URLLC problem as a Partially Observable Markov Decision Process (POMDP) and the introduction of an agent state, serving as a sufficient statistic of past observations and actions, enabling a transformation of the POMDP into a Markov Decision Process (MDP); 2) adapting the Proximal Policy Optimization (PPO) algorithm to handle the combinatorial action space; 3) incorporating prior knowledge into the learning agent with the introduction of a Bayesian policy. Numerical results reveal that not only does our approach outperform traditional multiple access protocols and DRL benchmarks on 3GPP scenarios, but also proves to be robust under various channel and traffic configurations, efficiently exploiting inherent time correlations., Comment: 35 pages, 11 figures, under review at the IEEE Transactions on Machine Learning in Communications and Networking journal

For decades, Simultaneous Ascending Auction (SAA) has been the most popular mechanism used for spectrum auctions. It has recently been employed by many countries for the allocation of 5G licences. Although SAA presents relatively simple rules, it induces a complex strategic game for which the optimal bidding strategy is unknown. Considering the fact that sometimes billions of euros are at stake in an SAA, establishing an efficient bidding strategy is crucial. In this work, we model the auction as a $n$-player simultaneous move game with complete information and propose the first efficient bidding algorithm that tackles simultaneously its four main strategic issues: the $\textit{exposure problem}$, the $\textit{own price effect}$, $\textit{budget constraints}$ and the $\textit{eligibility management problem}$. Our solution, called $SMS^\alpha$, is based on Simultaneous Move Monte Carlo Tree Search (SM-MCTS) and relies on a new method for the prediction of closing prices. By introducing a new reward function in $SMS^\alpha$, we give the possibility to bidders to define their own level of risk-aversion. Through extensive numerical experiments on instances of realistic size, we show that $SMS^\alpha$ largely outperforms state-of-the-art algorithms, notably by achieving higher expected utility while taking less risks., Comment: 20 pages, 19 figures, The paper has been submitted to IEEE journal for possible publication

The HST treasury program BUFFALO provides extended wide-field imaging of the six Hubble Frontier Fields galaxy clusters. Here we present the combined strong and weak-lensing analysis of Abell 370, a massive cluster at z=0.375. From the reconstructed total projected mass distribution in the 6arcmin x 6arcmin BUFFALO field-of-view, we obtain the distribution of massive substructures outside the cluster core and report the presence of a total of seven candidates, each with mass $\sim 5 \times 10^{13}M_{\odot}$. Combining the total mass distribution derived from lensing with multi-wavelength data, we evaluate the physical significance of each candidate substructure, and conclude that 5 out of the 7 substructure candidates seem reliable, and that the mass distribution in Abell 370 is extended along the North-West and South-East directions. While this finding is in general agreement with previous studies, our detailed spatial reconstruction provides new insights into the complex mass distribution at large cluster-centric radius. We explore the impact of the extended mass reconstruction on the model of the cluster core and in particular, we attempt to physically explain the presence of an important external shear component, necessary to obtain a low root-mean-square separation between the model-predicted and observed positions of the multiple images in the cluster core. The substructures can only account for up to half the amplitude of the external shear, suggesting that more effort is needed to fully replace it by more physically motivated mass components. We provide public access to all the lensing data used as well as the different lens models., Comment: 29 pages, 17 figures, 3 tables

The physical aging of supercooled glycerol induced by upward temperature steps of amplitude reaching 45 K was studied by a new method consisting in heating a micrometer-thick liquid film at a rate of up to 60 000 K/s, holding it at a constant high temperature for a controlled duration before letting it quickly cool down to the initial temperature. By monitoring the final slow relaxation of the dielectric loss, we were able to obtain quantitative information on the liquid response to the initial upward step. The so-called TNM (Tool-Narayanaswamy-Moynihan) formalism provided a good description of our observations despite the large distance from equilibrium, provided that different values of the nonlinearity parameter were used for the cooling phase and for the (much further from equilibrium) heating phase. In this form, it allowed to precisely quantify how to design an ideal temperature step, i.e., where no relaxation occurs during the heating phase. It helped bringing a clear physical understanding of how the (kilosecond long) final relaxation is related to the (millisecond long) liquid response to the upward step. Finally, it made possible the reconstruction of the fictive temperature evolution immediately following a step, evidencing the highly non-linear character of the liquid response to such large amplitude temperature steps. This work illustrates both the strengths and limitations of the TNM approach. This new experimental device offers a promising tool to study far-from-equilibrium supercooled liquids through their dielectric response., Comment: 11 pages, 9 figures and supplementary materials

We present a constraint on the abundance of supergiant (SG) stars at redshift z approx. 1, based on recent observations of a strongly lensed arc at this redshift. First we derive a free-form model of MACS J0416.1-2403 using data from the BUFFALO program. The new lens model is based on 72 multiply lensed galaxies that produce 214 multiple images, making it the largest sample of spectroscopically confirmed lensed galaxies on this cluster. The larger coverage in BUFFALO allows us to measure the shear up to the outskirts of the cluster, and extend the range of lensing constraints up to ~ 1 Mpc from the central region, providing a mass estimate up to this radius. As an application, we make predictions for the number of high-redshift multiply-lensed galaxies detected in future observations with JWST. Then we focus on a previously known lensed galaxy at z=1.0054, nicknamed Spock, which contains four previously reported transients. We interpret these transients as microcaustic crossings of SG stars and compute the probability of such events. Based on simplifications regarding the stellar evolution, we find that microlensing (by stars in the intracluster medium) of SG stars at z=1.0054 can fully explain these events. The inferred abundance of SG stars is consistent with either (1) a number density of stars with bolometric luminosities beyond the Humphreys-Davidson (HD) limit (L ~ $6\times10^5 L_{\odot}$) that is below 400 stars per sq. kpc, or (2) the absence of stars beyond the HD limit but with a SG number density of ~ 9000 per sq. kpc for stars with luminosities between $10^5$ and $6\times10^5$. This is equivalent to one SG star per 10x10 pc$^2$. We finally make predictions for future observations with JWST's NIRcam. We find that in observations made with the F200W filter that reach 29 mag AB, if cool red SG stars exist at z~1 beyond the HD limit, they should be easily detected in this arc, Comment: 24 pages & 18 figures

In biomedical imaging, deep learning-based methods are state-of-the-art for every modality (virtual slides, MRI, etc.) In histopathology, these methods can be used to detect certain biomarkers or classify lesions. However, such techniques require large amounts of data to train high-performing models which can be intrinsically difficult to acquire, especially when it comes to scarce biomarkers. To address this challenge, we use a single, pre-trained, deep embeddings extractor to convert images into deep features and train small, dedicated classification head on these embeddings for each classification task. This approach offers several benefits such as the ability to reuse a single pre-trained deep network for various tasks; reducing the amount of labeled data needed as classification heads have fewer parameters; and accelerating training time by up to 1000 times, which allows for much more tuning of the classification head. In this work, we perform an extensive comparison of various open-source backbones and assess their fit to the target histological image domain. This is achieved using a novel method based on a proxy classification task. We demonstrate that thanks to this selection method, an optimal feature extractor can be selected for different tasks on the target domain. We also introduce a feature space augmentation strategy which proves to substantially improve the final metrics computed for the different tasks considered. To demonstrate the benefit of such backbone selection and feature-space augmentation, our experiments are carried out on three separate classification tasks and show a clear improvement on each of them: microcalcifications (29.1% F1-score increase), lymph nodes metastasis (12.5% F1-score increase), mitosis (15.0% F1-score increase).

We present a new method to simultaneously/self-consistently model the mass distribution of galaxy clusters that combines constraints from strong lensing features, X-ray emission and galaxy kinematics measurements. We are able to successfully decompose clusters into their collisionless and collisional mass components thanks to the X-ray surface brightness, as well as using the dynamics of cluster members to obtain more accurate masses with the fundamental plane of elliptical galaxies. Knowledge from all observables is included through a consistent Bayesian approach in the likelihood or in physically motivated priors. We apply this method to the galaxy cluster Abell S1063 and produce a mass model that we publicly release with this paper. The resulting mass distribution presents a different ellipticities for the intra-cluster gas and the other large-scale mass components; and deviation from elliptical symmetry in the main halo. We assess the ability of our method to recover the masses of the different elements of the cluster using a mock cluster based on a simplified version of our Abell S1063 model. Thanks to the wealth of information provided by the mass model and the X-ray emission, we also found evidence for an on-going merger event with gas sloshing from a smaller infalling structure into the main cluster. In agreement with previous findings, the total mass, gas profile and gas mass fraction are consistent with small deviations from the hydrostatic equilibrium. This new mass model for Abell S1063 is publicly available as is the software used to construct it through the \textsc{Lenstool} package., Comment: 29 pages, 23 figures, Accepted for publication in MNRAS

The distribution of mass in galaxy-scale strong gravitational lenses is often modelled as an elliptical power law plus 'external shear', which notionally accounts for neighbouring galaxies and cosmic shear. We show that it does not. Except in a handful of rare systems, the best-fit values of external shear do not correlate with independent measurements of shear: from weak lensing in 45 Hubble Space Telescope images, or in 50 mock images of lenses with complex distributions of mass. Instead, the best-fit shear is aligned with the major or minor axis of 88% of lens galaxies; and the amplitude of the external shear increases if that galaxy is disky. We conclude that 'external shear' attached to a power law model is not physically meaningful, but a fudge to compensate for lack of model complexity. Since it biases other model parameters that are interpreted as physically meaningful in several science analyses (e.g. measuring galaxy evolution, dark matter physics or cosmological parameters), we recommend that future studies of galaxy-scale strong lensing should employ more flexible mass models., Comment: 13 pages, 11 figures, submitted to MNRAS and Review Response Submitted

Dirt cones are meter-scale structures encountered at the surface of glaciers, which consist of ice cones covered by a thin layer of ashes, sand or gravel, and which form naturally from an initial patch of debris. In this article, we report field observations of cone formation in the French Alps, laboratory-scale experiments reproducing these structures in a controlled environment, and two-dimensional discrete-element-method-finite-element-method numerical simulations coupling the grains mechanics and thermal effects. We show that cone formation originates from the insulating properties of the granular layer, which reduces ice melting underneath as compared to bare ice melting. This differential ablation deforms the ice surface and induces a quasi static flow of grains that leads to a conic shape, as the thermal length become small compared to the structure size. The cone grows until it reaches a steady state in which the insulation provided by the dirt layer exactly compensates for the heat flux coming from the increased external surface of the structure. These results allowed us to identify the key physical mechanisms at play and to develop a model able to quantitatively reproduce the various field observations and experimental findings., Comment: 10 pages, 8 figures, and supplementary materials (6 pages, 5 figures)

The automation of the medical evidence acquisition and diagnosis process has recently attracted increasing attention in order to reduce the workload of doctors and democratize access to medical care. However, most works proposed in the machine learning literature focus solely on improving the prediction accuracy of a patient's pathology. We argue that this objective is insufficient to ensure doctors' acceptability of such systems. In their initial interaction with patients, doctors do not only focus on identifying the pathology a patient is suffering from; they instead generate a differential diagnosis (in the form of a short list of plausible diseases) because the medical evidence collected from patients is often insufficient to establish a final diagnosis. Moreover, doctors explicitly explore severe pathologies before potentially ruling them out from the differential, especially in acute care settings. Finally, for doctors to trust a system's recommendations, they need to understand how the gathered evidences led to the predicted diseases. In particular, interactions between a system and a patient need to emulate the reasoning of doctors. We therefore propose to model the evidence acquisition and automatic diagnosis tasks using a deep reinforcement learning framework that considers three essential aspects of a doctor's reasoning, namely generating a differential diagnosis using an exploration-confirmation approach while prioritizing severe pathologies. We propose metrics for evaluating interaction quality based on these three aspects. We show that our approach performs better than existing models while maintaining competitive pathology prediction accuracy., Comment: Camera ready. NeurIPS 2022

We study copula-based collective risk models when the dependence structure is defined by a Farlie-Gumbel-Morgenstern (FGM) copula. By leveraging a one-to-one correspondence between the class of FGM copulas and multivariate symmetric Bernoulli distributions, we find convenient representations for the moments and Laplace-Stieltjes transform for the aggregate random variable defined from collective risk models with FGM dependence. We examine different components of this collective risk model, aiming to better understand the impact of the assumed dependence between a claim's frequency and severity. Relying on stochastic ordering, we analyze the impact of dependence on the aggregate claim amount random variable. Even if the FGM copula may only induce moderate dependence, we illustrate through numerical examples that the cumulative effect of FGM dependence can lead to substantial variations in key risk measures on aggregate random variables defined from collective risk models.

We propose an approach to construct a new family of generalized Farlie-Gumbel-Morgenstern (GFGM) copulas that naturally scales to high dimensions. A GFGM copula can model moderate positive and negative dependence, cover different types of asymmetries, and admits exact expressions for many quantities of interest such as measures of association or risk measures in actuarial science or quantitative risk management. More importantly, this paper presents a new method to construct high-dimensional copulas based on mixtures of power functions, and may be adapted to more general contexts to construct broader families of copulas. We construct a family of copulas through a stochastic representation based on multivariate Bernoulli distributions and Coxian-2 distributions. This paper will cover the construction of a GFGM copula, study its measures of multivariate association and dependence properties. We explain how to sample random vectors from the new family of copulas in high dimensions. Then, we study the bivariate case in detail and find that our construction leads to an asymmetric modified Huang-Kotz FGM copula. Finally, we study the exchangeable case and provide some insights into the most negative dependence structure within this new class of high-dimensional copulas.