Article
Physics, Fluids & Plasmas
Muneki Yasuda, Kaiji Sekimoto
Summary: The study proposes a method combining importance sampling and spatial Monte Carlo integration to evaluate expectations on Ising models. The method performs efficiently in both high- and low-temperature regions, effectively suppressing performance degradation in low-temperature regions.
Article
Computer Science, Software Engineering
Pascal Grittmann, Oemercan Yazici, Iliyan Georgiev, Philipp Slusallek
Summary: Multiple importance sampling (MIS) is a crucial tool in light-transport simulation, allowing robust Monte Carlo integration through the combination of samples from different techniques. However, the efficiency of complex combined estimators is not always superior to simpler algorithms, leading to the proposal of a general method to improve MIS efficiency.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Interdisciplinary Applications
DanHua ShangGuan
Summary: The Monte Carlo method is a powerful tool in many research fields, but the increasing complexity of physical models and mathematical models requires efficient algorithms to overcome the computational cost.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Physics, Multidisciplinary
Mateu Sbert, Laszlo Szirmay-Kalos
Summary: Multiple Importance Sampling (MIS) combines probability density functions (pdf) of various sampling techniques, optimizing weights to minimize variance. This paper discovers that MIS can be seen as a divergence problem between the integrand and the pdf, leading to simpler computations and more robust solutions.
Article
Engineering, Civil
John Thedy, Kuo-Wei Liao
Summary: A novel Importance Sampling method for calculating reliability in structural engineering problems using multiple spheres was proposed, which maximizes the number of safety samples by introducing spheres with different radii and centers, leading to reduced computational cost and improved efficiency. Compared to traditional methods, the proposed approach demonstrates higher robustness and efficiency.
Article
Engineering, Electrical & Electronic
Mateu Sbert, Laszlo Szirmay-Kalos
Summary: Multiple importance sampling combines probability density functions of multiple techniques into an importance function. The optimal combination weights are determined by solving a linear equation for a few initial samples, instead of numerically unstable variance optimization. The proposed method is validated with numerical examples and the direct lighting problem in computer graphics.
EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING
(2023)
Article
Mathematics, Applied
Tony Lelievre, Gabriel Stoltz, Wei Zhang
Summary: This study proposes a new MCMC algorithm for sampling probability distributions on submanifolds. Unlike previous methods, this algorithm uses set-valued maps in the proposal step and ensures the correctness of the sampling results through carefully enforced reversibility property.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2023)
Article
Physics, Multidisciplinary
Giulio Franzese, Dimitrios Milios, Maurizio Filippone, Pietro Michiardi
Summary: In this work, a novel practical method is proposed to make the sg noise isotropic by using a fixed learning rate determined analytically. Extensive experimental validations indicate that the proposal is competitive with the state of the art on sgmcmc.
Article
Physics, Multidisciplinary
Mateu Sbert, Victor Elvira
Summary: This paper proposes a novel and generic family of multiple importance sampling estimators. By revisiting the balance heuristic estimator and establishing a generalized framework, the authors study the optimal selection of free parameters and show that the resulting estimator has a lower variance than the balance heuristic estimator in most cases. Numerical examples demonstrate the improved efficiency of the new estimator compared to the classical balance heuristic.
Article
Computer Science, Theory & Methods
Yiolanda Englezou, Timothy W. Waite, David C. Woods
Summary: This study focuses on comparing various methods for approximating expected Shannon information gain in common nonlinear models, with ALIS being up to 70% cheaper than LIS and sometimes being both cheaper and more accurate.
STATISTICS AND COMPUTING
(2022)
Article
Computer Science, Software Engineering
Zackary Misso, Benedikt Bitterli, Iliyan Georgiev, Wojciech Jarosz
Summary: We introduce a general framework for transforming biased estimators into unbiased and consistent estimators in the same field. We demonstrate how this framework can be applied to rendering and improve existing unbiased estimation strategies. We provide examples of novel unbiased forms of transmittance estimation, photon mapping, and finite differences that are developed using this framework.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Engineering, Multidisciplinary
M. Croci, K. E. Willcox, S. J. Wright
Summary: This paper extends MLBLUE method to multi-output forward UQ problems and presents new semidefinite programming formulations for their optimal setup. These formulations yield the optimal number of samples required and the optimal selection of low-fidelity models to use. The new multi-output formulations can be solved reliably and efficiently.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Engineering, Electrical & Electronic
Victor Elvira, Ignacio Santamaria
Summary: In this paper, a multiple importance sampling (MIS) method is proposed for efficient symbol error rate (SER) estimation of maximum likelihood (ML) multiple input multiple output (MIMO) detectors. The method provides unbiased SER estimates by sampling from a mixture of components that are carefully chosen and parametrized, showing high performance in simulations with SERs lower than 10^(-8) accurately estimated with just 10^4 random samples.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2021)
Article
Engineering, Electrical & Electronic
Ali Mousavi, Reza Monsefi, Victor Elvira
Summary: Importance sampling is a powerful Monte Carlo methodology for approximating integrals, with adaptive importance sampling methods iteratively improving a set of proposals. The Hamiltonian Monte Carlo algorithm is popular in machine learning and statistics. The novel Hamiltonian adaptive importance sampling method achieves significant performance improvement in high-dimensional problems.
IEEE SIGNAL PROCESSING LETTERS
(2021)
Article
Engineering, Electrical & Electronic
Victor Elvira, Luca Martino, Pau Closas
Summary: This work presents a general framework of numerical integration techniques inspired by IS methodology, incorporating deterministic rules to reduce estimator errors and extending the applicability of Gaussian quadrature rules. The approach utilizes recent advances in multiple IS and adaptive literature, providing superior performance in various problems.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2021)
Article
Computer Science, Software Engineering
Arsene Perard-Gayot, Richard Membarth, Roland Leissa, Sebastian Hack, Philipp Slusallek
ACM TRANSACTIONS ON GRAPHICS
(2019)
Article
Physics, Applied
Bart van Dam, Benjamin Bruhn, Ivo Kondapaneni, Gejza Dohnal, Alexander Wilkie, Jaroslav Krivanek, Jan Valenta, Yvo D. Mudde, Peter Schall, Katerina Dohnalova
PHYSICAL REVIEW APPLIED
(2019)
Article
Computer Science, Software Engineering
Pascal Grittmann, Iliyan Georgiev, Philipp Slusallek, Jaroslav Krivanek
ACM TRANSACTIONS ON GRAPHICS
(2019)
Article
Computer Science, Software Engineering
Alexander Rath, Pascal Grittmann, Sebastian Herholz, Petr Vevoda, Philipp Slusallek, Jaroslav Krivanek
ACM TRANSACTIONS ON GRAPHICS
(2020)
Article
Crystallography
Patrick Trampert, Dmitri Rubinstein, Faysal Boughorbel, Christian Schlinkmann, Maria Luschkova, Philipp Slusallek, Tim Dahmen, Stefan Sandfeld
Summary: The study suggests a method for generating synthetic data based on parametric data modeling to enhance the generalization of trained neural network models. Particularly useful in situations where data collection is challenging, this approach may help in training neural networks effectively. The targeted data generation via adaptively sampling the parameter space of generative models yields better results compared to generating random data points.
Article
Computer Science, Software Engineering
Alexander Wilkie, Petr Vevoda, Thomas Bashford-Rogers, Lukas Hosek, Tomas Iser, Monika Kolarova, Tobias Rittig, Jaroslav Krivanek
Summary: The model significantly enhances visual realism compared to existing clear sky models by using scatterer distribution data from atmospheric measurements. It also introduces new features not found in fitted models, such as radiance patterns for post-sunset conditions.
ACM TRANSACTIONS ON GRAPHICS
(2021)
Article
Computer Science, Software Engineering
Pascal Grittmann, Oemercan Yazici, Iliyan Georgiev, Philipp Slusallek
Summary: Multiple importance sampling (MIS) is a crucial tool in light-transport simulation, allowing robust Monte Carlo integration through the combination of samples from different techniques. However, the efficiency of complex combined estimators is not always superior to simpler algorithms, leading to the proposal of a general method to improve MIS efficiency.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
Alexander Rath, Pascal Grittmann, Sebastian Herholz, Philippe Weier, Philipp Slusallek
Summary: Russian roulette and splitting are commonly used techniques to increase the efficiency of Monte Carlo estimators. This research proposes an iterative learning approach to determine optimal factors for Russian roulette and splitting during rendering, resulting in significant speed-ups in unidirectional path tracing.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Hardware & Architecture
Manuela Schuler, Richard Membarth, Philipp Slusallek
Summary: Efficient memory usage is crucial for training deep learning networks on resource-restricted devices. This article presents XEngine, an approach that schedules network operators to heterogeneous devices in low memory environments by determining checkpoints and recomputations of tensors, optimizing the end-to-end time.
ACM TRANSACTIONS ON ARCHITECTURE AND CODE OPTIMIZATION
(2022)
Article
Computer Science, Software Engineering
Pascal Grittmann, Iliyan Georgiev, Philipp Slusallek
Summary: Combining diverse sampling techniques through multiple importance sampling is crucial for robustness in modern Monte Carlo light transport simulation. The proposal of a correlation-aware heuristic, based on known path densities required for MIS, can achieve significantly lower error compared to the balance heuristic, without incurring additional computational and memory overhead.
COMPUTER GRAPHICS FORUM
(2021)