UQSay

Wednesday, June 3, 2026

UQSay #101

The hundred and first UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, June 11, 2026.

2–3 PM — Zonghao Chen (FSML, University College London)

Stationary MMD points

Approximation of a target probability distribution using a finite set of points is a problem of fundamental importance in numerical integration. Several authors have proposed to select points by minimising a maximum mean discrepancy (MMD), but the non-convexity of this objective typically precludes global minimisation. Instead, we consider the concept of stationary points of the MMD which, in contrast to points globally minimising the MMD, can be accurately computed. Our main contributions are two-fold and theoretical in nature. We first prove the (perhaps surprising) result that, for integrands in the associated reproducing kernel Hilbert space, the numerical integration error of stationary MMD points vanishes faster than the MMD. Motivated by this super-convergence property, we consider MMD gradient flows as a practical strategy for computing stationary points of the MMD. We then prove that MMD gradient flow can indeed compute stationary MMD points, based on a refined convergence analysis that establishes a novel non-asymptotic finite-particle error bound.

References:

Stationary MMD points, 2025

Joint work with Toni Karvonen (Lappeenranta–Lahti University of Technology) & Heishiro Kanagawa (Newcastle University) & François-Xavier Briol (UCL) & Chris Oates (Newcastle University).

Organizing committee: Pierre Barbillon (MIA-Paris), Julien Bect (L2S), Nicolas Bousquet (EDF R&D), Vincent Chabridon (EDF R&D), Amélie Fau (LMPS), Filippo Gatti (LMPS), Clément Gauchy (CEA), Bertrand Iooss (EDF R&D), Alexandre Janon (LMO), Sidonie Lefebvre (ONERA), Didier Lucor (LISN), Sébastien Petit (LNE), Emmanuel Vazquez (L2S), Xujia Zhu (L2S).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

If you want to attend this seminar (or any of the forthcoming online UQSay seminars), and if you do not already have access to the UQSay group on Teams, simply send an email and you will be invited. Please specify which email address the invitation must be sent to (this has to be the address associated with your Teams account).

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

The technical side of things: you can use Teams either directly from your web browser or using the "fat client", which is available for most platforms (Windows, Linux, Mac, Android & iOS). We strongly recommend the latter option whenever possible. Please give it a try before the seminar to anticipate potential problems.

Thursday, May 21, 2026

UQSay #100

The one hundredth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, May 28, 2026.

2–3 PM — Julien Bect & Xujia Zhu ( L2S, CentraleSupélec)

(Goal-Oriented) Global Sensitivity Analysis Revisited:The Mystery of the Camembert Slices

Sensitivity analysis plays a critical role in uncertainty quantification, aiming to characterize how uncertainty in model inputs propagates through computational models or experiments to the outputs. In contrast to local approaches, global methods account for variability over the entire input space, providing a more thorough description of input-output relationships. A wide range of global sensitivity indices has been proposed over the past decades, particularly to define so-called closed sensitivity indices , which quantify the joint contribution of a group (or coalition) of input variables.

From closed indices, one can easily derive first-order (main) effects and higher-order interaction effects. A desirable property in this setting is the non-negativity of the resulting sensitivity indices, which yields an interpretable decomposition of the total uncertainty, much like a pie chart---or ``Camembert'' diagram, as it is sometimes called in French---partitions a whole into non-overlapping contributions. However, this property does not hold in general. In fact, beyond the well-known case of variance-based (Sobol') sensitivity indices, to the best of our knowledge, only two frameworks ensure non-negative higher-order indices, both of them discovered quite recently by Da Veiga [1]: the first one relies on the expected Maximum Mean Discrepancy (MMD) between the conditional and the marginal distribution, while the second one leverages the Hilbert--Schmidt Independence Criterion (HSIC) in combination with specific (ANOVA) kernels.

In this talk, we first review three constructions of closed sensitivity indices available in the literature, in relation with the key notion of uncertainty functional [2]. Then we present a unified framework [3] that clarifies the common mechanism at work in the two classes of indices proposed by Da Veiga. At the heart of this framework resides a new avatar of the Sobol'-Hoeffding decomposition, also known as the functional ANOVA decomposition. Finally, we discuss several open questions and directions for future research, in particular regarding general necessary and sufficient conditions for higher-order indices to be non-negative.

References:

[1] Kernel-based ANOVA decomposition and Shapley effects—application to global sensitivity analysis, 2021
[2] Uncertainty functionals revisited: Concavity and Jensen’s inequality, 2026
[3] On non-negative coalitional decompositions in global sensitivity analysis, in preparation, 2026

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Tuesday, April 21, 2026

UQSay #99

The ninety-ninth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, May 7, 2026.

2–3 PM — Donatien Rossat ( EDF R&D)

Information Geometry-based Robust Bayesian Analysis

Bayesian inference provides a comprehensive framework for quantifying epistemic uncertainties, and updating them from new information. It relies on updating a so-called prior distribution, which summarizes the level of knowledge about some input parameters. In this work, we introduce a novel sensitivity analysis method to quantify the influence of prior distributions on Bayesian inference outcomes. We define perturbed-law-based sensitivity indices (PLI), which measure the effect of uncertainties in prior specification through controlled perturbations of a reference prior. These perturbations are constructed using the Fisher distance from information geometry, enabling a consistent exploration of a wide range of deviations beyond infinitesimal changes. We further show that these indices can be reformulated as relative variations of rare event probabilities, allowing efficient computation using existing reliability methods. The proposed approach is illustrated on Bayesian inverse problems of varying complexity. Results demonstrate its ability to identify parameters for which prior choices significantly impact Bayesian inference results, while remaining applicable to nonlinear and high-dimensional settings.

References:

Information Geometry-based Robust Bayesian Analysis, International Journal for Uncertainty Quantification , 2026

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Thursday, April 2, 2026

UQSay #98

The ninety-eighth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, April 9, 2026.

2–3 PM — Lucia Clarotto (MIA Paris-Saclay, AgroParisTech - Geolearning Chair) — [slides]

Prediction of spatio-temporal Gaussian processes by advection-diffusion stochastic partial differential equations

In the task of predicting spatio-temporal fields in environmental science using statistical methods, introducing statistical models inspired by the physics of the underlying phenomena that are numerically efficient is of growing interest. Large space–time datasets call for new numerical methods to efficiently process them. The Stochastic Partial Differential Equation (SPDE) approach has proven to be effective for the estimation and the prediction in a spatial context. We present here the advection–diffusion SPDE with first–order derivative in time which defines a large class of nonseparable spatio-temporal models, both on Euclidean spaces and Riemamnnian manifolds. A Gaussian Markov random field approximation of the solution to the SPDE is built by discretizing the temporal derivative with a finite difference method and by solving the spatial SPDE with a finite element method at each time step. Computationally efficient methods are proposed to estimate the parameters of the SPDE and to predict the spatio-temporal field by kriging, as well as to perform conditional simulations. The approach is applied to datasets of solar radiation and atmospheric aerosol optical depth across the Earth’s surface.

References:

The SPDE approach for spatio-temporal datasets with advection and diffusion, Spatial Statistics, 2024
Prediction of spatio-temporal data on meshed surfaces using advection-diffusion SPDEs, 2025

Joint work with Denis Allard (INRAE) & Nicolas Desassis & Mike Pereira & Thomas Romary (Mines PSL)

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Thursday, March 19, 2026

UQSay #97

The ninety-seventh UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, March 26, 2026.

2–3 PM — Anthony Quintin (CEA DIF)

Optimal experimental designs based on the cross-entropy method for planning fracture toughness tests

Nuclear reactor pressure vessels undergo progressive embrittlement under neutron irradiation, monitored through the Master Curve theory which estimates a reference temperature T0. The core challenge is obtaining a reliable estimate of T0 from a very limited number of specimens. The work pursues two objectives. First, developing a numerical twin of fracture toughness test campaigns based on finite element simulations coupled with a Beremin model calibrated via Bayesian inference. Second, building a decision-making tool to optimize experimental planning within the framework of Bayesian Optimal Experimental Design, which seeks to determine the experimental conditions maximizing the information gained from a limited number of observations.

To this end, the approach relies on a Bayesian optimization method aimed at identifying test temperature sequences that minimize uncertainty on T0. The problem is formulated as a constrained combinatorial optimization, where the criterion to maximize is an expected information gain (entropy). Temperature sequences are modeled as homogeneous first-order discrete-time Markov chains, whose transition matrix is optimized via the Cross-Entropy Method. The resulting method yields a transition matrix that serves as a directly interpretable sequential decision rule. The proposed methodology is inherently general and transferable to any material and to a broad range of experimental testing frameworks where observations are limited and acquisition costs are high.

References:

Joint work with Jean Marc Bourinet & Cécile Mattrand (SIGMA Clermont) & Rudy Chocat (CEA Saclay) & Tom Petit (CEA Gramat)

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Thursday, February 26, 2026

UQSay #96

The ninety-sixth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, March 12, 2026.

2–3 PM — Richard Everitt (Department of Statistics, University of Warwick) — [slides]

Improved MCMC with active subspaces

Constantine et al. (2016) introduced a Metropolis-Hastings (MH) approach that target the active subspace of a posterior distribution: a linearly projected subspace that is informed by the likelihood.. Schuster et al. (2017) refined this approach to introduce a pseudo-marginal Metropolis-Hastings, integrating out inactive variables through estimating a marginal likelihood at every MH iteration. In this talk we show empirically that the effectiveness of these approaches is limited in the case where the linearity assumption is violated, and suggest a particle marginal Metropolis-Hastings algorithm as an alternative for this situation. The high computational cost of these approaches leads us to consider alternative approaches to using active subspaces in MCMC that avoid the need to estimate a marginal likelihood: we introduce Metropolis-within-Gibbs and Metropolis-within-particle Gibbs methods that provide a more computationally efficient use of the active subspace.

References:

Improved MCMC with active subspaces, 2025

Joint work with Leonardo Ripoli (University of Reading)

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, February 4, 2026

UQSay #95

The ninety-fifth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, February 12, 2026.

2–3 PM — Yuansi Chen (ETH Zürich) — [slides]

When does Metropolized Hamiltonian Monte Carlo provably outperform Metropolis-adjusted Langevin algorithm?

We analyze the mixing time of Metropolized Hamiltonian Monte Carlo (HMC) with the leapfrog integrator to sample from a distribution on $\mathbb{R}^d$ whose log-density is smooth, has Lipschitz Hessian in Frobenius norm and satisfies isoperimetry. We bound the gradient complexity to reach $\epsilon$ error in total variation distance from a warm start by $O(d^{1/4} \polylog(1/\epsilon))$ and compare it to the minimax mixing rate of MALA. We discuss the benefit of the leapfrog integrator in HMC with short integration time.

References:

Joint work with Khashayar Gatmiry (MIT) & Minhui Jiang (ETH Zürich)

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Thursday, January 15, 2026

UQSay #94

The ninety-fourth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, ~~January 22~~ → February 05 (rescheduled), 2026.

2–3 PM — Eleni Chatzi (ETH Zürich, Department of Civil, Environmental and Geomatic Engineering) — [slides]

Dynamics, Inference, and Uncertainty: Foundations of AI-Enhanced Digital Twins

This talk examines the role of structural dynamics as a foundation for inference under uncertainty in AI-enhanced Digital Twins. Structural assets are highly individual, operate under variable environmental and operational conditions, and are only partially observable. In this context, dynamical signatures, extracted through structured inference schemes such as modal analysis, provide compact and physically interpretable representations through which uncertainty can be quantified, propagated, and reduced.

The talk discusses how physics-enhanced machine learning (PEML), together with structured representations and reduced-order models, enables potent yet efficient approximations that synergize with available data. By embedding physical constraints, governing equations, and dynamical structure into learning architectures, these models balance prior knowledge with data-driven adaptation, reducing epistemic uncertainty while remaining scalable under limited or evolving observations. Structured low-dimensional representations further support stable learning, efficient uncertainty propagation, and interpretable model updates.

By framing dynamics as the interface between physics and data within the context of inference, the talk highlights pathways toward uncertainty-aware, interpretable digital twins capable of supporting resilient decision-making of complex infrastructure systems.

References:

Joint work with the further members of the group of Structural Mechanics & Monitoring at ETH Zürich

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Friday, December 19, 2025

UQSay #93

The ninety-third UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, January 8, 2026.

2–3 PM — Masha Naslidnyk (Department of Statistical Science, University College London) — [slides]

Kernel Quantile Embeddings and Associated Probability Metrics

Embedding probability distributions into reproducing kernel Hilbert spaces (RKHS) has enabled powerful non-parametric methods such as the maximum mean discrepancy (MMD), a statistical distance with strong theoretical and computational properties. At its core, the MMD relies on kernel mean embeddings (KMEs) to represent distributions as mean functions in RKHS. However, it remains unclear if the mean function is the only meaningful RKHS representation. Inspired by generalised quantiles, we introduce the notion of kernel quantile embeddings (KQEs), along with a consistent estimator. We then use KQEs to construct a family of distances that: (i) are probability metrics under weaker kernel conditions than MMD; (ii) recover a kernelised form of the sliced Wasserstein distance; and (iii) can be efficiently estimated with near-linear cost. Through hypothesis testing, we show that these distances offer a competitive alternative to MMD and its fast approximations. Our findings demonstrate the value of representing distributions in Hilbert space beyond simple mean functions, paving the way for new avenues of research.

References:

Kernel Quantile Embeddings and Associated Probability Metrics, ICML 2025

Joint work with Siu Lun Chau (NTU, Singapore) & François-Xavier Briol (UCL) & Krikamol Muandet (CISPA Helmholtz Center for Information Security).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, December 10, 2025

UQSay #92

The ninety-second UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, December 18, 2025.

2–3 PM — Lea Friedli (Engineering Risk Analysis Group, Technical University of Munich)

CRPS-Based Targeted Sequential Design with Application in Chemical Space

Gaussian processes (GPs) have become a widely used tool for modeling unknown functions across various domains. In many applications, particular interest lies in a specific range of the response, with the goal of identifying inputs that lead to desired outputs. To enhance GP model performance in this setting, we employ weighted scoring rules to develop sequential design strategies that selectively augment the training dataset. Specifically, we study pointwise and integral criteria based on the threshold-weighted Continuous Ranked Probability Score (CRPS), using two different weighting measures. We showcase an application in synthetic chemistry, where the objective is to identify molecules with specific properties. However, the presented acquisition strategies are applicable to a wide range of fields and pave the way to further developing sequential design relying on scoring rules.

References:

CRPS-Based Targeted Sequential Design with Application in Chemical Space, 2025 - github

Joint work with Athénaïs Gautier (ONERA) & Anna Broccard (OFJ) & David Ginsbourger (IMSV, University of Bern).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, November 26, 2025

UQSay #91

The ninety first UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, December 4, 2025.

2–3 PM — Iain Henderson (ISAE-SUPAERO) — [slides]

Multidimensional conformal prediction with random ellipsoids

Conformal prediction (CP) is a popular framework for performing uncertainty quantification on a given a statistical predictor. Its main perks are as follow : (i) little to no assumption on the distribution of the data is required, and (ii) CP provides finite sample coverage garanties. The CP confidence regions are built using a so-called ''conformity score'', which dictates the properties of the said regions. In this talk, I will describe two new conformity scores in a general multivariate regression framework. They are based on a covariance analysis of the residuals and the input points. I will provide theoretical guarantees on the prediction sets, which consist in explicit ellipsoids. We study the asymptotic properties of the ellipsoids, and show that their volume is reduced compared to that of classic balls, under ellipticity assumptions. I will provide numerical illustrations of our results, including heavy-tailed as well as non-elliptical distributions.

References:

Adaptive inference with random ellipsoids through Conformal Conditional Linear Expectation, 2024

Joint work with Adrien Mazoyer & Fabrice Gamboa (Institut de Mathématiques de Toulouse).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, November 12, 2025

UQSay #90

The ninetieth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, November 20, 2025.

2–3 PM — Fanny Lehmann (ETH AI Center, ETH Zurich)

Foundation Models of the Earth System: Seeing Beyond Weather

Deep learning has revolutionized weather forecasting over the past three years, with AI models surpassing the accuracy of traditional numerical simulations at a fraction of the computational cost. In this talk, I will present how these models—and specifically, foundation models—can be extended beyond weather forecasting. I will show that the latent space of foundation models is sufficiently rich to predict new physical variables with minimal, lightweight fine-tuning. I will also explore the conditions under which some foundation models remain indefinitely stable for long autoregressive predictions, challenging the common belief that such models inevitably accumulate errors to the point of blow-up. These findings open new perspectives for applying AI models to climate projections and quantify uncertainties in climate change scenarios.

References:

Finetuning a Weather Foundation Model with Lightweight Decoders for Unseen Physical Processes, 2025

Joint work with the SwissAI Initiative team for Weather and Climate.

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, October 22, 2025

UQSay #89

The eighty-ninth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, October 30, 2025.

2–3 PM — Edgar Jaber (EDF R&D, Centre Borelli, LISN) — [slides]

A Bayesian methodology for hybrid degradation prognostics

Degradation prognostics of industrial assets involves estimating their remaining useful life (RUL) by projecting current health indicators and operating conditions while quantifying associated uncertainties. These prognostics are central to the development and deployment of digital twins, which aim to provide insights into the evolving state of complex systems. Traditionally, RUL estimation relies on physics-based simulations or data-driven models. While both have their merits, they can prove inadequate when simulation runtimes are prohibitive or when degradation data is sparse, common challenges in digital twin implementations for critical industrial infrastructure.

To address this problem, we developed an offline modular data assimilation approach. Firstly, a Bayesian model updating strategy combines kernel-based sensitivity analysis to identify and rank the time-varying influence of the model’s input variables, with a tailored inference scheme that accounts for the heterogeneity of available data. Posterior distributions are sampled using MCMC techniques, while the method mitigates the curse of dimensionality by iteratively updating the marginals of influential input variables under an independence assumption. Posterior informativeness is quantified through the Kullback–Leibler divergence, comparing updated distributions to their priors. Secondly, the full state distribution is updated with the help of an ensemble Kalman smoothing step, further reducing the posterior uncertainty.

After detailing the methodology, I will illustrate how this approach enhances the fidelity of RUL predictions and reduces uncertainty in a clogging prognostics use case for digital twins of steam generators in nuclear power plants.

References:

Fusion of heterogeneous data for robust degradation prognostics, 2025 - github

Joint work with Emmanuel Remy (EDF R&D) & Vincent Chabridon (EDF R&D) & Mathilde Mougeot (ENS Paris-Saclay) & Didier Lucor (LISN).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Monday, October 6, 2025

UQSay #88

The eighty-eighth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, October 16, 2025.

2–3 PM — Virginie Ehrlacher (CERMICS, Ecole Nationale des Ponts et Chaussées) — [slides]

Marginal-constrained modified Wasserstein barycenters for Gaussian distributions and Gaussian mixtures

The aim of this talk is to present some modified Wasserstein barycenters for probability measures defined on cartesian product sets which satisfy given marginal constraints. We focus on the specific case of Gaussian and Gaussian mixture distributions, as the proposed approach strongly relies on new results about properties of geometric means of covariance matrices. In the case of Gaussian distributions, the marginal-constrained modified Wasserstein barycenters can be analytically computed, while for Gaussian mixtures, computing the marginal-constrained barycenter consists in a postprocessing of the Gaussian mixture Wasserstein barycenter. In both cases, we provide numerical simulations illustrating the difference between Wasserstein barycenters and modified marginal-constrained Wasserstein barycenters. We moreover provide several test cases where the marginal-constrained Wasserstein barycenters interpolate better than regular Wasserstein barycenters, showcasing the practical interest of the proposed approach. As a by-product, we prove new results concerning marginal-preserving Wasserstein barycenters. Indeed, Wasserstein barycenters do not preserve marginals in general. In this work, as a consequence of the derived properties on the geometric mean of covariance matrices, we obtain sufficient and necessary conditions for the Wasserstein barycenter between two Gaussian distributions to preserve marginals, and provide necessary conditions in the case of more than two Gaussians.

References:

Marginal-constrained modified Wasserstein barycenters for Gaussian distributions and Gaussian mixtures, 2025

Joint work with Maxime Daléry (LMB - Université Franche-Comté) & Geneviève Dusson (LMB - Université Franche-Comté).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, September 3, 2025

UQSay #87

The eighty-seventh UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, September 18, 2025.

2–3 PM — Pamphile Roy (LUT Business School, LUT University - Consulting Manao GmbH, Austria) — [slides]

A novel way of visualizing causal uncertainty

Global sensitivity analysis (GSA) is crucial for understanding model behavior and informing decision-making. However, its adoption is hindered by methodological complexity, implementation challenges, and high computational costs. To address these issues, we developed Simulation Decomposition (SimDec), a hybrid approach that simplifies GSA through efficient computation of variance-based sensitivity indices and intelligent visualization techniques. SimDec is made accessible to practitioners of any background via a no-code web dashboard. The latest enhancement to the SimDec dashboard includes two-output graphs, which allow users to visualize relationships between two model outputs alongside their marginal distributions. This feature is demonstrated through a case study on optimizing a heat exchanger in a nuclear reactor, examining the relationship between the levelized cost of heat and mechanical design characteristics. By providing an intuitive, no-code interface, SimDec democratizes GSA, making it accessible to users with limited mathematical training. This work was presented at the SAMO 2025 conference, highlighting the potential of SimDec to transform how sensitivity analysis is conducted across various fields.

References:

Uncovering heterogeneous effects in computational models for sustainable decision-making, Environmental Modelling & Software 2024
Simulation Decomposition in Python, Journal of Open Source Software 2024 - [SimDec dashboard]
Simple binning algorithm and SimDec visualization for comprehensive sensitivity analysis of complex computational models, Journal of Environmental Informatics Letters 2025

Joint work with Mariia Kozlova (LUT University) & Andrea Saltelli (UPF Barcelona School of Management) & Julian Scott Yeomans (York University).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Tuesday, May 6, 2025

UQSay #86

The eighty-sixth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, May 15, 2025.

2–3 PM — Johanna Ziegel (ETH Zürich) — [slides]

(Conformal) isotonic distributional regression

Isotonic distributional regression (IDR) is a nonparametric distributional regression approach under a monotonicity constraint. It has found application as a generic method for uncertainty quantification, in statistical postprocessing of weather forecasts, and in distributional single index models. IDR has favorable in-sample calibration and optimality properties, which allow to conformalize it and obtain out-of-sample online guarantees.

References:

Isotonic Distributional Regression, JRSS Series B 2023
Easy Uncertainty Quantification (EasyUQ): Generating Predictive Distributions from Single-Valued Model Output, SIAM Review 2024
In-sample calibration yields conformal calibration guarantees, 2025

Joint work with S. Allen (KIT) & G. Gavrilopoulos & A. Henzi (ETH Zürich) & T. Gneiting & E-M. Walz (HITS-KIT).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, March 26, 2025

UQSay #85

The eighty-fifth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, April 3, 2025.

2–3 PM — Bilel Bensaid (Toulouse School of Economics) — [slides]

New insights in neural networks optimization: Lyapunov stability and splitting schemes

These recent years, a great number of algorithms have been developed to optimize neural networks parameters (p-GD, clipping GD, Momentum, RMSProp, Adam, ...) but they need an accurate tuning to be stable and efficient. To get rid of the long and experimental step of GridSearch, we are looking for adaptive optimizers that come with guarantees. By analysing the stability of these algorithms, a general methodology to adapt the learning rate is suggested (generalization of the Armijo rule) for any deep learning optimizers, relating "robust" optimizers to preserving discretization schemes. Convergence and complexity of these methods are discussed leading to acceleration results, promoting the use of adaptive learning rate strategies for Analytic and Recurrent Neural Networks.

Finally, this study is extended to the mini-batch setting, revealing the link between mini-batch optimization and splitting operator methods. In a nutshell, this work comes up with deep relations between neural network training and classical issues in the numerical analysis of differential equations. .

References:

Joint work with G. Poette (CEA DAM, CESTA - ENSEIRB-Matmeca) & R. Turpault (IMB - ENSEIRB-Matmeca).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, February 26, 2025

UQSay # 84

The eighty-fourth UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, March 6, 2025.

2–3 PM — Elena Di Bernardino (Laboratoire Jean Alexandre Dieudonné, Université Côte d'Azur) — [slides]

Curvature measures for random excursion sets: theoretical and computational developments

The excursion set of a smooth random field carries relevant information in its various geometric measures. Geometric properties of these exceedance regions above a given level provide meaningful theoretical and statistical characterizations for random fields defined on Euclidean domains. Many theoretical results have been obtained for excursions of Gaussian processes and include expected values of the so-called Lipschitz-Killing curvatures (LKCs), such as the area, perimeter and Euler characteristic in two-dimensional Euclidean space. In this talk we will describe a recent series of theoretical and computational contributions in this field. Our aim is to provide answers to questions like:

- How the geometric measures of an excursion set can be inferred from a discrete sample of the excursion set;

- How these measures can be related back to the distributional properties of the random field from which the excursion set was obtained;

- How the excursion set geometry can be used to infer the extremal behavior of random fields.

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Tuesday, February 4, 2025

UQSay #83

The eighty-third UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, February 13, 2025.

2–3 PM — Margot Hérin (LIP6, Sorbonne University) — [slides]

Algorithms for learning capacity-based preference models

Preference models from Decision Theory are used to describe, explain, or predict human behavior in evaluation or decision-making tasks. Beyond this descriptive role, a key feature of these models is their ability to guarantee normative properties, ensuring the internal consistency of the modeled value system and the resulting decisions. Hence, they can also be used to assist individuals in making a relevant choice based on their preferences or provide machines with the ability to autonomously yet controllably make sophisticated decisions in complex environments, involving multi-criteria or collective decision-making, or decision-making under uncertainty.

In this talk, we consider aggregation functions weighted by a non-additive set function (called capacity), such as multilinear utilities or Choquet integrals. The non-additivity of the capacity makes it possible to model criteria interactions, leaving room for a diversity of attitudes in criteria aggregation. However, allowing for these interactions dramatically increases the complexity of the preference learning task and may prevent the model from being interpretable, due to the combinatorial nature of the possible interactions.

We address this challenge by learning a sparse Möbius transform of the capacity, where the few non-zero Möbius masses indicate the significant positive or negative synergies between criteria. Specifically, we propose a learning method based on iterative reweighted least squares (IRLS) for sparse recovery, and dualization to improve scalability, making it possible to handle aggregation problems involving more than 20 criteria. We also present an online learning algorithm based on regularized dual averaging (RDA), designed for decision-making contexts where preference examples become available sequentially, but also well suited to handle large-scale preference databases (large number of preferences or criteria examples). In addition, the inclusion of normative constraints on the capacity (e.g., monotonicity, supermodularity) is made possible by combining RDA with the method of alternating direction multipliers (ADMM).

References:

Learning Preference Models with Sparse Interactions of Criteria, IJCAI 2023
Online Learning of Capacity-Based Preference Models, IJCAI 2024

Joint work with P. Perny (Sorbonne University) & N. Sokolovska (Sorbonne University).

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Wednesday, January 22, 2025

UQSay #82

The eighty-second UQSay seminar on UQ, DACE and related topics will take place online on Thursday afternoon, January 30, 2025.

2-3 PM — Alexandros A. Taflanidis (University of Notre Dame, Department of Civil and Environmental Engineering and Earth Sciences, USA) — [slides]

Reduced order and surrogate modeling applications for computationally efficient uncertainty propagation within seismic vulnerability assessment

Seismic vulnerability assessment involves the quantification and propagation of the different courses of uncertainty impacting structural performance. For engineering demand parameters (EDPs) the relevant uncertainties pertain to the seismic hazard and/or to the structural model characteristics, while the detailed characterization of structural vulnerability is typically performed using nonlinear response-history analysis (NLRHA). Despite recent advances in computational science, the adoption of computationally intensive, high-fidelity finite element models (FEMs) for performing NLRHA remains a challenge for many seismic risk assessment applications, forcing some sort of simplification of the uncertainty characterization. This presentation will investigate two alternative computational statistics approaches for improving computational efficiency in such settings.

The first approach will be the use of reduced order models (ROMs), coupled, if needed, with a Multi-Fidelity Monte Carlo (MFMC) implementation. ROMs simplify the physics-based description of the original FEM through some form of condensation of the degrees of freedom and equations of motion, coupled with an approximation of the nonlinear (hysteretic) response characteristics. In order to accommodate any potential bias from the ROM approximation, a MFMC setting is additionally examined. In the latter setting, the ROM serves as a means to accelerate the Monte Carlo convergence, relying ultimately on the FEM to establish unbiased predictions.

The second approach will be the use of surrogate models, offering an entirely data-driven mathematical approximation of the input/output relationships of the high-fidelity model. For addressing aleatoric uncertainties in the hazard description (i.e., the so-called ground-motion to ground-motion variability), a stochastic Gaussian Process (GP) emulation approach is adopted to directly approximate the EDP distribution (considering the influence of the aleatoric uncertainties). Improvements in computational efficiency are promoted by avoiding any replications for the stochastic GP implementation. The extension to vector EDP outputs is also briefly discussed, accommodated by approximating the correlation matrix.

Coordinators: Sidonie Lefebvre (ONERA) & Xujia Zhu (L2S)

Practical details: the seminar will be held online using Microsoft Teams.

You will find the link to the seminar on the "General" UQSay channel on Teams, approximately 15 minutes before the beginning.

Pages

Wednesday, June 3, 2026

2–3 PM — Zonghao Chen (FSML, University College London)

Stationary MMD points

Thursday, May 21, 2026

2–3 PM — Julien Bect & Xujia Zhu ( L2S, CentraleSupélec)

(Goal-Oriented) Global Sensitivity Analysis Revisited:The Mystery of the Camembert Slices

Tuesday, April 21, 2026

2–3 PM — Donatien Rossat ( EDF R&D)

Information Geometry-based Robust Bayesian Analysis

Thursday, April 2, 2026

2–3 PM — Lucia Clarotto (MIA Paris-Saclay, AgroParisTech - Geolearning Chair) — [slides]

Prediction of spatio-temporal Gaussian processes by advection-diffusion stochastic partial differential equations

Thursday, March 19, 2026

2–3 PM — Anthony Quintin (CEA DIF)

Optimal experimental designs based on the cross-entropy method for planning fracture toughness tests

Thursday, February 26, 2026

2–3 PM — Richard Everitt (Department of Statistics, University of Warwick) — [slides]

Improved MCMC with active subspaces

Wednesday, February 4, 2026

2–3 PM — Yuansi Chen (ETH Zürich) — [slides]

When does Metropolized Hamiltonian Monte Carlo provably outperform Metropolis-adjusted Langevin algorithm?

Thursday, January 15, 2026

2–3 PM — Eleni Chatzi (ETH Zürich, Department of Civil, Environmental and Geomatic Engineering) — [slides]

Dynamics, Inference, and Uncertainty: Foundations of AI-Enhanced Digital Twins

Friday, December 19, 2025

2–3 PM — Masha Naslidnyk (Department of Statistical Science, University College London) — [slides]

Kernel Quantile Embeddings and Associated Probability Metrics

Wednesday, December 10, 2025

2–3 PM — Lea Friedli (Engineering Risk Analysis Group, Technical University of Munich)

CRPS-Based Targeted Sequential Design with Application in Chemical Space

Wednesday, November 26, 2025

2–3 PM — Iain Henderson (ISAE-SUPAERO) — [slides]

Multidimensional conformal prediction with random ellipsoids

Wednesday, November 12, 2025

2–3 PM — Fanny Lehmann (ETH AI Center, ETH Zurich)

Foundation Models of the Earth System: Seeing Beyond Weather

Wednesday, October 22, 2025

2–3 PM — Edgar Jaber (EDF R&D, Centre Borelli, LISN) — [slides]

A Bayesian methodology for hybrid degradation prognostics

Monday, October 6, 2025

2–3 PM — Virginie Ehrlacher (CERMICS, Ecole Nationale des Ponts et Chaussées) — [slides]

Marginal-constrained modified Wasserstein barycenters for Gaussian distributions and Gaussian mixtures

Wednesday, September 3, 2025

2–3 PM — Pamphile Roy (LUT Business School, LUT University - Consulting Manao GmbH, Austria) — [slides]

A novel way of visualizing causal uncertainty

Tuesday, May 6, 2025

2–3 PM — Johanna Ziegel (ETH Zürich) — [slides]

(Conformal) isotonic distributional regression

Wednesday, March 26, 2025

2–3 PM — Bilel Bensaid (Toulouse School of Economics) — [slides]

New insights in neural networks optimization: Lyapunov stability and splitting schemes

Wednesday, February 26, 2025

2–3 PM — Elena Di Bernardino (Laboratoire Jean Alexandre Dieudonné, Université Côte d'Azur) — [slides]

Curvature measures for random excursion sets: theoretical and computational developments

Tuesday, February 4, 2025

2–3 PM — Margot Hérin (LIP6, Sorbonne University) — [slides]

Algorithms for learning capacity-based preference models

Wednesday, January 22, 2025

2-3 PM — Alexandros A. Taflanidis (University of Notre Dame, Department of Civil and Environmental Engineering and Earth Sciences, USA) — [slides]

Reduced order and surrogate modeling applications for computationally efficient uncertainty propagation within seismic vulnerability assessment