Ting Wang, U.S. Army Research Laboratory
Jaroslaw Knap, U.S. Army Research Laboratory
Petr Plechac, University of Delaware
Gideon Simpson, Drexel University
Many advances in science and engineering are dependent upon mathematical models and their computer simulation. These numerical models aim to capture behavior of complex physical systems with many degrees of freedom, complex dynamics, and multiple time scales. Because of that, purely deterministic models often do not adequately span the problem space. There are many reasons why deterministic models yield inadequate results. For example, a system may be described by well-established physical laws, but certain model parameters cannot be determined experimentally with sufficient accuracy. Alternatively, the interaction of a system with its environment may involve a great number of rapidly varying forces with no predictable pattern. A system with many degrees of freedom may also require a large number of complex and expensive experiments in order to properly describe interactions between system components. Finally, laws governing the behavior of a system may not be fully known. Stochastic modeling is critical for situations, like those above, when deterministic models fail to provide a sufficient or relevant description. Yet, owing to inherent complexities associated with stochastic modeling, novel computational methods are vital in order to overcome these challenges. In addition, recent advances in data science have enabled data-enhanced stochastic methods that take advantage of large-scale data.
In this minisymposium, we bring together researchers and practitioners to review some recent advances on data-enhanced stochastic computational methods with a focus on modeling of physical systems. Key topics will include, but are not limited to:
- Monte Carlo sampling methods
- Uncertainty quantification of complex systems
- Stochastic simulation/optimization
- Physics-informed machine learning with provable guarantees
- Novel stochastic modeling in science and engineering