Meeting Summary: This three-day workshop will bring together mathematicians, statisticians, computational scientists, computer scientists, and application domain experts across science, engineering and medicine to address the topic of scientific machine learning for complex systems, with a focus on moving beyond forward simulation to achieve inference and optimization at scale.
Scientific machine learning is a growing field that brings together the complementary perspectives of computational science and computer science to craft a new generation of machine learning methods for complex applications across science and engineering. In these applications, dynamics are complex and multiscale, data are sparse and expensive to acquire, decisions have high consequences, and uncertainty quantification is essential. Furthermore, applications often demand predictions that go well beyond the available data. The goal is not just to model these systems, but to learn the models from data, and optimize the control, design, or operation of these systems, all under uncertainty. Scientific machine learning recognizes the potential benefit of new ideas in the field of machine learning, but also emphasizes the essential role of mathematical and physics-based models in incorporating predictive power, interpretability, and domain knowledge.
This workshop will cover topics at the forefront of applied mathematics and computational science research to achieve scientific machine learning in complex systems with a focus on moving “beyond forward simulation.” Particular challenges and topics will include:
• the tight and dynamic interplay between computational models and physical systems
• decision making and control on rapid time scales
• assimilation of multiple heterogeneous data sources
• multiphysics and multiscale coupling and emergent behavior
• surrogate and reduced-order modeling
• the need for robustness to data and model uncertainties to support high consequence decisions
• design optimization with costly computational models
• predictive digital twins