Bayesian Inverse Problems for EUV reflectometry with regularized JKO scheme

DOI10.48550/ARXIV.2611.09574MaRDI QIDQ7433560arXivFDO

Authors Vitalii Aksenov, Oliver Henze

Publication date 6 November 2026

Published in arXiv

Tldr We propose a novel Bayesian inverse problem solver, based on an Eulerian JKO scheme that uses proximal steps in Wasserstein space and efficiently learns high-dimensional densities from EUV reflectometry data, leveraging tensor-train format for scalability. We demonstrate its effectiveness on a ruthenium-silicon specimen and show it can be reused to obtain samples for shifted target distributions with less computational effort than MCMC.

Full work available at URL https://doi.org/10.48550/ARXIV.2611.09574

arXiv classification

math.NA

Cites work

• A computational fluid mechanics solution to the Monge-Kantorovich mass transfer problem
• A continuous analogue of the tensor-train decomposition
• A kernel formula for regularized Wasserstein proximal operators
• A Simplex Method for Function Minimization
• An Eulerian approach to regularized JKO scheme with low-rank tensor decompositions for Bayesian inversion
• Anderson acceleration for fixed-point iterations
• Application of accelerated fixed-point algorithms to hydrodynamic well-fracture coupling
• Approximation and sampling of multivariate probability distributions in the tensor train decomposition
• Breaking the Curse of Dimensionality, Or How to Use SVD in Many Dimensions
• Conditional Wasserstein distances with applications in Bayesian OT flow matching
• Constructive representation of functions in low-rank tensor formats
• Convergence of entropic schemes for optimal transport and gradient flows
• emcee: The MCMC Hammer
• Ensemble samplers with affine invariance
• Entropic approximation of Wasserstein gradient flows
• Fisher information regularization schemes for Wasserstein gradient flows
• Importance sampling: intrinsic dimension and computational cost
• Iterative importance sampling algorithms for parameter estimation
• Kronecker products and matrix calculus: with applications
• Mean-field Variational Inference via Wasserstein Gradient Flow
• Model-Based Derivative-Free Optimization Methods and Software
• On Information and Sufficiency
• On the limited memory BFGS method for large scale optimization
• Optimal metropolis algorithms for product measures on the vertices of a hypercube
• Optimization with PDE Constraints
• preCICE -- a fully parallel library for multi-physics surface coupling
• Proximal splitting methods in signal processing
• Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distances
• Solution of the Fokker-Planck equation by cross approximation method in the tensor train format
• Stochastic processes and applications. SPAS2017, Västerås and Stockholm, Sweden, October 4--6, 2017
• Tensor networks for dimensionality reduction and large-scale optimization. I: Low-rank tensor decompositions
• Tensor Networks for Latent Variable Analysis. Part I: Algorithms for Tensor Train Decomposition
• Tensor-train decomposition
• The Variational Formulation of the Fokker--Planck Equation
• Trust Region Methods
• TT-cross approximation for multidimensional arrays

This page was built for publication: Bayesian Inverse Problems for EUV reflectometry with regularized JKO scheme

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7433560)