A5015839438- Probabilistic and Robust Engineering Design
- Gaussian Processes and Bayesian Inference
- Advanced Multi-Objective Optimization Algorithms
- Structural Health Monitoring Techniques
- Scientific Research and Discoveries
- Model Reduction and Neural Networks
- Image and Signal Denoising Methods
- Image Processing and 3D Reconstruction
- Fault Detection and Control Systems
- Statistical Methods and Inference
- Statistical Methods and Bayesian Inference
- Bayesian Methods and Mixture Models
- Spectroscopy and Quantum Chemical Studies
- Reservoir Engineering and Simulation Methods
- Chaos control and synchronization
- Sparse and Compressive Sensing Techniques
- Protein Structure and Dynamics
- Nonlinear Photonic Systems
- Nonlinear Dynamics and Pattern Formation
- Markov Chains and Monte Carlo Methods
- Blind Source Separation Techniques
- Machine Learning and Algorithms
- Topology Optimization in Engineering
- Fractional Differential Equations Solutions
- Soil Geostatistics and Mapping
GE Vernova (United States)
2025
General Electric (United States)
2020-2023
General Electric (Israel)
2021-2023
GE Global Research (United States)
2021
École Polytechnique Fédérale de Lausanne
2016-2020
University of Southern California
2012-2019
Czech Academy of Sciences, Institute of Mathematics
2018
Experimental design is crucial for inference where limitations in the data collection procedure are present due to cost or other restrictions. Optimal experimental designs determine parameters that some appropriate sense make most informative possible. In a Bayesian setting this translated updating best possible posterior. Information theoretic arguments have led formation of expected information gain as criterion. This can be evaluated mainly by Monte Carlo sampling and maximized using...
A novel approach is presented for constructing polynomial chaos representations of scalar quantities interest (QoI) that extends previously developed methods adaptation in Homogeneous Chaos spaces. In this work, we develop a Bayesian formulation the problem characterizes posterior distributions series coefficients and rotation matrix acting on Gaussian input variables. The thus construed as new parameter map from to QoI, estimated through inference. For computation coefficients'...
Abstract A new method is proposed for efficient optimization under uncertainty that addresses the curse of dimensionality as it pertains to evaluation probabilistic objectives and constraints. basis adaptation strategy previously introduced by authors integrated into a design framework construes cost function quantity interest computes stochastic adapted bases functions space parameters. With these bases, integrations at each point are evaluated low-dimensional integrals (mostly one...
The major drawback of the Bayesian approach to model calibration is computational burden involved in describing posterior distribution unknown parameters arising from fact that typical Markov chain Monte Carlo (MCMC) samplers require thousands forward evaluations. In this work, we develop a variational which uses an information theoretic criterion recast problem as optimization problem. Specifically, parameterize using family Gaussian mixtures and seek minimize loss incurred by replacing...
Abstract Gaussian process (GP) regression or kriging has been extensively applied in the engineering literature for purposes of building a cheap-to-evaluate surrogate, within contexts multi-fidelity modeling, model calibration, and design optimization. With ongoing automation manufacturing industrial practices as part Industry 4.0, there greater need advancing GP techniques to handle challenges such high input dimensionality, data paucity big problems, these consist primarily proposing...
Parametric dictionaries can increase the ability of sparse representations to meaningfully capture and interpret underlying signal information, such as encountered in biomedical problems. Given a mapping function from atom parameter space actual atoms, we propose Bayesian framework for learning parameters, because its provide full posterior estimates, take uncertainty into account generalize on unseen data. Inference is performed with Markov Chain Monte Carlo, that uses block sampling...
Uncertainty propagation and sensitivity analysis in expensive engineering design problems require meticulous selection of the experimental data sophisticated analytic methods, order to leverage maximum amount information that is carried within. Although a metamodeling approach typically suffices as way provide cheap-to-evaluate generative model, classical surrogate methods often fail accurately describe high dimensional outputs. In this work we attempt tackle challenging problem airfoil...
Abstract The recently introduced basis adaptation method for homogeneous (Wiener) chaos expansions is explored in a new context where the rotation/projection matrices are computed by discovering active subspace (AS) random input exhibits most of its variability. In case One-dimensional (1D) AS exists, methodology can be applicable to generalized polynomial (PCE), thus enabling projection high-dimensional single variable and efficient estimation univariate expansion. Attractive features this...
In the present work, we examine combined effects of cubic and quintic terms long-range type in dynamics a double well potential. Employing two-mode approximation, systematically develop two cubic–quintic ordinary differential equations assess contributions interactions each relevant prefactors, gauging how to simplify ensuing dynamical system. Finally, obtain reduced canonical description for conjugate variables relative population imbalance phase between wells proceed systems analysis...
Inverse problems in engineering are of critical importance for efficient design and calibration various industrial operation maintenance settings. Capturing correctly the model uncertainties by characterizing distributions conditional on observation data is a common task that has been addressed extensively literature. In this work, we developing flexible framework solving not single, but family inverse more specifically exploring marginal posterior as well only observations also or other...
Bayesian techniques for engineering problems, which rely on Gaussian process (GP) regression, are known their ability to quantify epistemic and aleatory uncertainties being data efficient. The mathematical elegance of applying these methods usually comes at a high computational cost when compared deterministic empirical methods. Furthermore, using becomes practically infeasible in scenarios characterized by large number inputs thousands training data. focus this work is enhancing based...
In this work, an efficient uncertainty quantification approach is presented to estimate the likelihood of successful delivery a payload by sailplane in presence uncertainties its flight and wing conditions. A probabilistic treatment both aleatory interval-defined epistemic variables employed here - allowing be versatile enough handle mixed types input variables. particular, variables, often defined through interval-bounds, are translated representation modeling them via family continuous...
Solving inverse problems in scientific and engineering fields has long been intriguing holds great potential for many applications, yet most techniques still struggle to address issues such as high dimensionality, nonlinearity model uncertainty inherent these problems. Recently, generative models Generative Adversarial Networks (GANs) have shown approximating complex dimensional conditional distributions paved the way characterizing posterior densities Bayesian problems, problems'...