HAL-Polytechnique
Not a member yet
51406 research outputs found
Sort by
Apprentissage profond contraint pour l'évaluation et la couverture d'options européennes en marchés incomplets
No full textIn incomplete financial markets, pricing and hedging European options lack a unique no-arbitrage solution due to unhedgeable risks. This paper introduces a constrained deep learning approach to determine option prices and hedging strategies that minimize the Profit and Loss (P&L) distribution around zero. We employ a single neural network to represent the option price function, with its gradient serving as the hedging strategy, optimized via a loss function enforcing the self-financing portfolio condition. A key challenge arises from the non-smooth nature of option payoffs (e.g., vanilla calls are non-differentiable atthe-money, while digital options are discontinuous), which conflicts with the inherent smoothness of standard neural networks. To address this, we compare unconstrained networks against constrained architectures that explicitly embed the terminal payoff condition, drawing inspiration from PDE-solving techniques. Our framework assumes two tradable assets: the underlying and a liquid call option capturing volatility dynamics. Numerical experiments evaluate the method on simple options with varying non-smoothness, the exotic Equinox option, and scenarios with market jumps for robustness. Results demonstrate superior P&L distributions, highlighting the efficacy of constrained networks in handling realistic payoffs. This work advances machine learning applications in quantitative finance by integrating boundary constraints, offering a practical tool for pricing and hedging in incomplete markets.Dans les marchés financiers incomplets, l'évaluation et la couverture des options européennes ne possèdent pas de solution unique sans arbitrage. Cet article propose une approche d'apprentissage profond contraint pour déterminer les prix des options et les stratégies de couverture qui minimisent la distribution du Profit and Loss (P&L) autour de zéro. Nous utilisons un seul réseau de neurones pour représenter la fonction de prix de l'option, son gradient servant de stratégie de couverture, le tout optimisé via une fonction de perte qui impose la condition d’auto-financement du portefeuille. Un défi majeur provient de la nature non régulière des valeurs d'exercice d'options (par exemple, les calls vanilles ne sont pas différentiables au prix d'exercice, tandis que les options binaires ne sont pas continues), ce qui entre en conflit avec la régularité naturelle des réseaux de neurones standards. Pour y remédier, nous comparons des réseaux non contraints à des architectures contraintes qui intègrent explicitement la condition de gain terminal, en s’inspirant des techniques de résolution d'EDP. Notre cadre suppose deux actifs négociables : le sous-jacent et une option call liquide capturant la dynamique de volatilité. Les expériences numériques évaluent la méthode sur des options simples présentant différents degrés d'irrégularité, sur l’option exotique Equinoxe, ainsi que dans des scénarios avec sauts de marché pour tester la robustesse .Les résultats montrent des distributions de P&L nettement supérieures, démontrant l’efficacité des réseaux contraints pour traiter des payoffs réalistes. Ce travail fait progresser les applications de l’apprentissage automatique en finance quantitative en intégrant des contraintes de conditions aux limites, et propose un outil pratique pour l'évaluation et la couverture dans les marchés incomplets
Immunomodulatory Biomimetic Collagen Scaffolds for Enhanced Tissue Regeneration
No full textInternational audienceIntroduction Understanding the dynamic interplay between cells and their microenvironment is fundamental to developing effective pro-regenerative tissue engineering strategies. However, there is still little knowledge on the cues involved in tissue repair. Our research focuses on identifying and isolating key extracellular matrix (ECM) signals involved in tissue regeneration to construct a biomimetic ECM model that fosters repair. Restoring the endothelium or epithelium is critical for the functional integration of engineered vascular and tracheal grafts. In vivo, the basal lamina (BL), primarily composed of laminin (Ln), fibronectin (Fn), and collagen IV (ColIV), lines the luminal surfaces of arteries and the trachea, providing structural and biochemical support for endothelial and epithelial cell adhesion. Additionally, regulatory T cells (Tregs) have emerged as key mediators of tissue regeneration, with their role in epithelial reconstruction being a topic of interest. Given that interleukin-2 (IL-2) is essential for the expansion and survival of CD4+ T cells, including Tregs, IL-2-loaded biomaterials may represent a promising strategy to accelerate tissue repair.Here we propose functionalized collagen-based scaffolds as models to investigate the effect of extracellular matrix and immune molecules on cell behavior for vascular and tracheal tissue regeneration, with evaluation of the pro-regenerative potential of collagen, basal lamina-proteins and anti-inflammatory cytokines. Materials and Methods Flat type I collagen scaffolds were fabricated using ice-templating followed by topotactic fibrillogenesis to closely replicate the collagen concentration and organization within the interstitial ECM (1,2). To evaluate their capacity to support the formation of an endothelial cell monolayer, scaffolds were functionalized with biomimetic coatings Ln, Fn, or ColIV, before seeding with bovine aortic endothelial cells (BAECs).In parallel, uncoated scaffolds were supplemented with varying concentrations of interleukin-2 (IL-2) to investigate their immunomodulatory potential. IL-2 stability and loading capacity by impregnation were assessed via ELISA. The biological activity of IL-2-loaded scaffolds was then evaluated on CD4+ T cells cultured in complete medium containing CD3/CD28 co-stimulatory signals. Cell viability, proliferation, and phenotype were analyzed by flow cytometry, data analysis was performed using FlowJo software. Results Endothelial cells successfully adhered to the surface of our biomimetic matrix, regardless of the coating, and expressed VE-cadherin, a key junction protein required for the formation of a tight monolayer (Figure 1). Moreover, the presence of biomimetic coatings significantly enhanced BAECs' metabolic activity on flat surfaces. Notably, polar orientation analysis revealed that Fn coating promoted cell alignment on these substrates.In optimized conditions, various amounts of IL2 could be impregnated inside 6 mm diameter collagen constructs, by modulating the concentration of the impregnation bath. After 6 days of CD4+ T cells culture, IL-2-loaded scaffolds exhibited a dose-dependent effect on Treg stimulation and proliferation within the total CD4+ population (Figure 2). Second Harmonic Generation (SHG) imaging confirmed that CD4+ T cells infiltrated the type I collagen matrix within six days. Furthermore, flow cytometry demonstrated that infiltrated cells proliferated more than those in suspension or in the control group, even in IL-2-free scaffolds. Discussion Cell-matrix interactions drive specific cellular responses, which can be modulated by modifying the substrate's composition. Our findings indicate that BAEC orientation on Fn coatings and T cell expansion within the collagen network are likely mediated by adhesion molecules involved in matrix-cell junctions and soluble mediators of cell-cell communication. The observed dose-dependent immunomodulatory effect of IL-2-loaded scaffolds on cells in suspension is directly linked to IL-2 availability in the surrounding medium, depending on IL2 diffusion from the substrate. A substantial portion of IL-2 is expected to remain trapped within the collagen network, which may sustain prolonged immunostimulation and enhance Treg recruitment by chemoattraction. Conclusions These findings open an exciting pathway to characterize and harness cell-ECM interactions, by modulating fibrous protein content and organisation, but also through incorporation of immunomodulatory soluble factors, such as pro-regenerative IL2 cytokine. This approach opens new perspectives for designing advanced biomaterials that support both structural and immunological aspects of tissue regeneration.(1) Rieu C. et al, 2019, 10.1021/acsami.9b03219 (2) Martinier I. et al., 2024, 10.1039/d3bm01808
Quantile Regression, Variational Autoencoders, and Diffusion Models for Uncertainty Quantification: A Spatial Analysis of Sub-seasonal Wind Speed Prediction
No full textInternational audienceThis study aims to improve the spatial representation of uncertainties when regressing surface wind speeds from large-scale atmospheric predictors for sub-seasonal forecasting. Sub-seasonal forecasting often relies on large-scale atmospheric predictors such as 500 hPa geopotential height (Z500), which exhibit higher predictability than surface variables and can be downscaled to obtain more localised information. Previous work by Tian et al. (2024) demonstrated that stochastic perturbations based on model residuals can improve ensemble dispersion representation in statistical downscaling frameworks, but this method fails to represent spatial correlations and physical consistency adequately. More sophisticated approaches are needed to capture the complex relationships between large-scale predictors and local-scale predictands while maintaining physical consistency. Probabilistic deep learning models offer promising solutions for capturing complex spatial dependencies. This study evaluates three probabilistic methods with distinct uncertainty quantification mechanisms: Quantile Regression Neural Network that directly models distribution quantiles, Variational Autoencoders that leverage latent space sampling, and Diffusion Models that utilise iterative denoising. These models are trained on ERA5 reanalysis data and applied to ECMWF sub-seasonal hindcasts to regress probabilistic wind speed ensembles. Our results show that probabilistic downscaling approaches provide more realistic spatial uncertainty representations compared to simpler stochastic methods, with each probabilistic model offering different strengths in terms of ensemble dispersion, deterministic skill, and physical consistency. These findings establish probabilistic downscaling as an effective enhancement to operational sub-seasonal wind forecasts for renewable energy planning and risk assessment
Heavy-Tailed Diffusion with Denoising Lévy Probabilistic Models
No full textInternational audienceExploring noise distributions beyond Gaussian in diffusion models remains an open challenge. While Gaussian-based models succeed within a unified SDE framework, recent studies suggest that heavy-tailed noise distributions, like -stable distributions, may better handle mode collapse and effectively manage datasets exhibiting class imbalance, heavy tails, or prominent outliers. Recently, Yoon et al.\ (NeurIPS 2023), presented the Lévy-Itô model (LIM), directly extending the SDE-based framework to a class of heavy-tailed SDEs, where the injected noise followed an -stable distribution, a rich class of heavy-tailed distributions. However, the LIM framework relies on highly involved mathematical techniques with limited flexibility, potentially hindering broader adoption and further development. In this study, instead of starting from the SDE formulation, we extend the denoising diffusion probabilistic model (DDPM) by replacing the Gaussian noise with -stable noise. By using only elementary proof techniques, the proposed approach, Denoising Lévy Probabilistic Models (DLPM), boils down to vanilla DDPM with minor modifications. As opposed to the Gaussian case, DLPM and LIM yield different training algorithms and different backward processes, leading to distinct sampling algorithms. These fundamental differences translate favorably for DLPM as compared to LIM: our experiments show improvements in coverage of data distribution tails, better robustness to unbalanced datasets, and improved computation times requiring smaller number of backward steps
Styx in Action: Transactional Cloud Applications Made Easy
No full textInternational audienceDeveloping and deploying transactional cloud applications such as banking and e-commerce systems is a daunting task for developers. The reason for this difficulty is twofold. First, developing such applications shifts the developers' focus from the application logic to considerations of distributed transactions, fault-tolerance, consistency, and scalability. Second, deploying such applications involves multiple systems, such as databases, load balancers, or containerized services, impeding efficient resource management. This demonstration presents Styx, a scalable application runtime that allows developers to build scalable and transactional cloud applications with minimal effort. It supports serializability and exactly-once guarantees and focuses on the ease of development and deployment, as well as Styx's fault-tolerance mechanisms
Styx: Transactional Stateful Functions on Streaming Dataflows
No full textInternational audienceDeveloping stateful cloud applications, such as low-latency workflows and microservices with strict consistency requirements, remains arduous for programmers. The Stateful Functions-as-a-Service (SFaaS) paradigm aims to serve these use cases. However, existing approaches provide weak transactional guarantees or perform expensive external state accesses requiring inefficient transactional protocols that increase execution latency. In this paper, we present Styx, a novel dataflow-based SFaaS runtime that executes serializable transactions consisting of stateful functions that form arbitrary call-graphs with exactly-once guarantees. Styx extends a deterministic transactional protocol by contributing: i) a function acknowledgment scheme to determine transaction boundaries required in SFaaS workloads, ii) a function-execution caching mechanism, and iii) an early commit-reply mechanism that substantially reduces transaction execution latency. Experiments with the YCSB, TPC-C, and Deathstar benchmarks show that Styx outperforms state-of-the-art approaches by achieving at least one order of magnitude higher throughput while exhibiting near-linear scalability and low latency
A Comparative Study of News Exposure and Consumption On and Off Facebook
No full textInternational audienceSocial media giants like Meta, Google, and X leverage powerful algorithms to personalize user feeds, a practice now under intense public scrutiny. These algorithms can inadvertently skew the information users consume, potentially influencing political opinions and voting decisions. This raises critical questions: Do social media platforms foster misinformation and contribute to echo chambers? To address this ongoing debate, our study directly compares news exposure on Facebook (where algorithmic influence is strong) with news consumption off-platform (where user behavior plays a larger role). Specifically, we investigate: (1) Are users exposed to more/less misinformation on Facebook compared with their off-platform misinformation consumption? (2) Is news exposure on Facebook more/less diverse than off-platform news consumption? (3) To what extent do socio-demographic and psychological factors influence misinformation exposure on Facebook and consumption off Facebook? (4) Is there a relationship between socio-demographic and psychological factors and news diversity on and off Facebook? and (5) Is users' exposure to misinformation on Facebook correlated to off-platform news consumption? The longstanding biggest barrier to answering these questions has been the lack of access to data on what information users see and consume while browsing the Internet. In this paper, we use a measurement approach that asks a panel of users to donate data about the content they see online. For this, we designed a tool to collect traces of all news articles that individuals encounter on their desktop Facebook timeline and while they browse the Internet (off Facebook), along with signals about how users interact with them (e.g., clicks, time spent reading). Our tool observes content and interactions on and off Facebook on 4,149 news media domains sourced from Media Bias Fact Check and NewsGuard. Alongside the news post and article collection, we conduct surveys to gather socio-demographic and psychological data from our participants. Our study of 123,995 news-related posts on Facebook and 70,587 news articles visits off Facebook, collected from 642 users during 12 weeks, reveals the following central findings: (1) Only a small fraction 4% of users' news consumption off Facebook is driven by news exposure on Facebook, and only 5.7% of misinformation consumption off Facebook is driven by news exposure on Facebook. (2) There is a higher prevalence of misinformation in user-received content on Facebook compared to deliberately consumed content off-platform. On Facebook, 5.9% of our users' news exposure comes from sources known for spreading misinformation, while off-platform, only 2.6% of our users' news consumption is from misinformation sources. Conversely, Facebook presents more diverse content - 22% of users received content from only one political leaning on Facebook, compared to 36% of users who consumed content from only one political leaning off-platform. (3) Several socio-demographic and psychological factors showed a statistically significant correlation with misinformation exposure on Facebook but not misinformation consumption off Facebook. (4) The proportion of misinformation consumed off Facebook emerged as a statistically significant predictor of users' exposure to misinformation on Facebook, independent of news consumption on Facebook
Ligands iminophosphorane polydentes avec des métaux abondants pour la catalyse moléculaire
No full textThe objective of this PhD. project was to synthetize original electron-rich ligands containing iminophosphorane moieties and evaluate their potential to develop earth abundant metal (3d) complexes for applications in molecular catalysis.Initially, a thorough experimental study paired with theoretical calculations (calculations done by Prof. A. Monari) investigated the electronic properties of bidentate iminophosphorane ligands, especially the influence of the nitrogen’s substituents on electron donation of this function. Four bidentate iminophosphorane-phosphine ligands with different substituents on the nitrogen (i.e. isopropyl, phenyl, trimethylsilane or hydrogen groups) were synthetized for this purpose. They were then coordinated to palladium(II) precursors to form square planar diamagnetic complexes. Introducing one or two isonitrile probes into the complexes allowed for the appraisal of the donation of the iminophosphoranes. This study underlined the importance of the choice of the nitrogen’s substituent on the electron donation potential of iminophosphoranes. Later, this was exemplified in the synthesis of different tridentate iminophosphorane ligands for molecular catalysis applications.A series of three neutral NNN ligands featuring an iminophosphorane (with three different phosphine groups: triethyl, triphenyl or tricyclohexyl phosphine), a central amine and a pyridine, was designed and coordinated to iron(II), cobalt(II) and manganese(II). Theses complexes were all proven active in the catalytic transfer hydrogenation of ketones with low catalytic loadings (1 mol%). Given the scarcity of examples of CoII and MnII catalysts for this reaction, we focused our study on these complexes. They were efficient on a large scope of ketones. Despite the paramagnetism of the involved species, some additional experiments were conducted to get an insight into the mechanism of this catalytic reaction. They suggest a metal-hydride complex as the active catalyst.In parallel, an original anionic ONP ligand bearing a hydroxy moiety, a central iminophosphorane and a phosphine was synthetized and was successfully coordinated to nickel(II) metal precursors. These complexes were very efficient in the catalytic hydrosilylation of olefins (with an anti-Markovnikov selectivity) and even ketones, with low catalytic loadings (1 mol%), at room temperature in 1 hour. This contrasts with literature precedents where very few nickel(II) complexes exhibit comparable activity to convert ketones. Furthermore, this catalyst is to our knowledge the only one able to selectively target ketones in a mixed olefin-ketone substrate. Some experiments suggest the involvement of a nickel-hydride as active catalyst. DFT calculations (done by Prof. V. Gandon) were very helpful in understanding the possible intermediates of this reaction, in particular by suggesting a distortion of the ligand into a butterfly shape, which allows the insertion of the substrate.The synthesis of tridentate ligands with a central phosphorus unit was also explored. A neutral NPN ligand with a central phosphine moiety and two lateral iminophosphoranes was built-up and coordinated to cobalt(II). This cobalt complex was shown to be catalytically active in the hydrosilylation of internal olefins with low catalytic loadings (1.5 mol%), at 60 °C within 6 hours, which is amongst the best performances reported until now.This PhD project is an illustrative specimen of the potential of using earth-abundant iminophosphorane complexes for catalytic applications.L’objectif de ce doctorat a été la synthèse de ligands novateurs comportant la fonction iminophosphorane électrodonneuse et d’évaluer leur potentiel pour développer des complexes de métaux abondants (3d) pour des applications en catalyse moléculaire.Dans un premier temps, une étude expérimentale et théorique (calculs réalisés par le Prof. A. Monari) des propriétés d’électro-donation des iminophosphoranes et de l’impact du substituant de l’azote sur celles-ci a été effectuée. Pour ce faire, une série de quatre ligand bidentes phosphine-iminophosphorane différent par le substituant de l’atome d’azote (isopropyl, phenyl, trimethylsilane ou hydrogène) a été synthétisée. Ces ligands ont été coordonnés au palladium(II) pour conduire à des complexes plan carré diamagnétiques. La donation des iminophosphoranes a été évaluée avec l’introduction d’une ou de deux sondes isonitriles dans ces complexes. Ces études ont permis de souligner l’importance du choix des substituants portés par l’azote sur l’électro-donation importante des iminophosphoranes. Ultérieurement, ces informations ont été mises en pratique dans la synthèse de divers ligands tridentes contenant des iminophosphoranes pour des applications en catalyse.Une série de trois ligands NNN neutres avec des fonctions : iminophosphoranes (différant par le substituant du phosphore : éthyle, phényle, ou cyclohexyle), une amine centrale, et une pyridine a été préparée. Ces ligands ont été coordonnés à des précurseurs de fer(II), cobalt(II) et manganèse(II). Ces complexes se sont montrés capables de catalyser l’hydrogénation par transfert de cétones avec des taux catalytiques bas (1 mol%). Compte tenu du peu d’exemples dans la littérature de catalyseurs à base de CoII et MnII, nous nous sommes focalisés sur ces derniers. Ils sont capables de réaliser l’hydrogénation d’une large gamme de cétones. Malgré les difficultés inhérentes au paramagnétisme des espèces mises en jeu, des expériences supplémentaires laissent penser que le catalyseur actif serait une espèce métal-hydrure.Un ligand anionique ONP possédant une structure originale de par la présence des fonctions phénolate, iminophosphorane en position centrale et phosphine a été conçu et coordonné à des précurseurs de nickel(II). Les complexes formés sont très efficaces dans l’hydrosilylation catalytique d’oléfines (sélectivité anti-Markovnikov) et de cétones avec un taux catalytique bas (1 mol%), à température ambiante en 1 heure. Les complexes de nickel sont des catalyseurs connus d’hydrosilylation d’oléfines, mais, il y a très peu d’exemples aussi efficaces pour la conversion des cétones. De plus, ce catalyseur est à notre connaissance, le seul à sélectivement convertir des cétones dans des substrats mixtes comportant des fonctions oléfines et cétones. Les tests complémentaires réalisés semblent indiquer que le catalyseur actif serait un complexe nickel-hydrure. Les calculs DFT (réalisés par Prof. V. Gandon) ont permis de proposer des intermédiaires réactionnels. Notamment, ces calculs suggèrent la distorsion du ligand sous une forme de papillon qui permet l’insertion du substrat.La synthèse de ligand iminophosphorane tridentes comportant une unité centrale phosphorée a également été explorée. Un ligand neutre NPN associant une phosphine centrale et deux iminophosphorane latéraux a été obtenu et coordonné au cobalt(II). Le complexe résultant permet l’hydrosilylation d’oléfines internes avec un taux catalytique bas (1,5 mol %), à 60 °C après 6 heures, ce qui est très correct au regard de ce qui est connu dans la littérature.Ce doctorat est donc une illustration du potentiel des complexes de métaux abondants à ligands iminophosphorane en catalyse moléculaire
Data Work in Egypt: Who Are the Workers Behind Artificial Intelligence?
No full textThe report highlights the role of Egyptian data workers in the global value chains of Artificial Intelligence (AI). These workers generate and annotate data for machine learning, check outputs, and they connect with overseas AI producers via international digital labor platforms, where they perform on-demand tasks and are typically paid by piecework, with no long-term commitment.Most of these workers are young, highly educated men, with nearly two-thirds holding undergraduate degrees. Their primary motivation for data work is financial need, with three-quarters relying on platform earnings to cover basic necessities. Despite the variability in their online earnings, these are generally low, often equaling Egypt's minimum wage. Data workers' digital identities are shaped by algorithmic control and economic demands, often diverging from their offline selves. Nonetheless, they find ways to resist, exercise ethical agency, and maintain autonomy.The report evaluates the potential impact of Egypt's newly enacted labor law and suggests policy measures to improve working conditions and acknowledge the role of these workers in AI's global value chains