You are here

Machine Learning for Solving Inverse Problems in Computational Mechanics and Materials

Jiaxin Zhang, ORNL

Ahn Tran, Sandia National Laboratories

Zhen Hu, University of Michigan-Dearborn

Tim Wildey, Sandia National Laboratories

This minisymposium focuses on computational inverse problems using state-of-the-art machine learning techniques. It is a highly multidisciplinary field of great importance with applications in computational mechanics and materials, including constitutive model calibration and advanced materials design. However, many inverse problems are known to be grand challenging due to e.g., high-dimensional feature spaces, complicated or unspecified models, limited observed data, stochasticity, and computationally intensive physical solver or simulator. Recent advances in machine learning provide alternatives owing to its several ingredients: (1) deep neural networks as a flexible framework for representing high-dimensional and complicated functions and maps, (2) generative models such as variational autoencoder and generative adversarial networks enable to project the high dimensionality to a low-dimensional latent space, and (3) advanced optimization algorithms with physical constraints.  It is natural to leverage the recent developments of machine learning in the study of inverse problems in the context of physics.

We are particularly interested in research that involves inverse modeling and design problems in computational mechanics and materials.  We invite researchers to submit work particularly in the following and related areas:
•  Computational inverse modeling problems
•  Advanced inverse materials design and discovery
•  Stochastic inverse problems in mechanics, materials, and manufacturing
•  Probabilistic machine learning methods
•  Deep learning including neural networks and generative models
•  Advanced optimization algorithms  
•  Machine learning application to uncertainty quantification
•  Advanced statistical methods (Monte Carlo method, Bayesian inference, etc.)