Nikolaos Bouklas, Cornell University
WaiChing Sun, Columbia University
Krishna Garikipati, University of Michigan
Deep learning and data-driven approaches have gained significant popularity in the field of computer science and engineering mechanics as alternatives to more conventional modeling approaches. Nevertheless, to fulfill the promise of accelerating simulation/digital twin workflow and providing feasible engineering solutions, it is necessary to ensure robustness and trustworthiness, especially for models intended to predict high-consequent engineering problems (e.g. aircraft design, structural engineering of building, nuclear waste disposal).
This mini-symposium welcomes contributions and new ideas that improve the feasibility, robustness, reliability, and accuracy of the data-driven/AI modeling approach for constitutive laws of solid materials that exhibit elastic and path-dependent behaviors. Possible contributions may include but are not limited to innovations on (1) handling limited labeled data (e.g. stress-strain curves), (2) robust validating, (3) intelligent data generation, exploration, and curation, (4) handling of multi-task constraints and Pareto-optimal solution under uncertainty, (5) causality based interpretability and causal discovery, (6) meta-learning and transfer learning methods that ensure the robustness of the neural network training as well as (7) uncertainty quantification and (8) integration of the knowledge of mechanics and physics towards machine learning approaches for engineering applications.