Dr. Stephen DeWitt is a computational materials scientist at the US Department of Energy’s Oak Ridge National Laboratory (ORNL). Before joining ORNL he was at the University of Michigan, where he received a BSE in Engineering Physics, a PhD in Applied Physics, and then held a research faculty position in the Department of Materials Science and Engineering. His research centers on integrating scalable, GPU-accelerated simulation tools into scientific workflows. He is particularly interested in the combination of multiscale simulations, in-situ characterization, artificial intelligence, optimization, and uncertainty quantification to drive innovation in materials and manufacturing.
Presentation Title:
Autonomous Decision-Making to Power Next-Generation Scientific Experiments
Presentation Abstract:
Today, too many scientific workflows are bottlenecked by steps performed by human scientists. Often these steps are mundane, and recent efforts toward automated workflows can eliminate them. However, some bottlenecks in these workflows involve decisions made using intuition and domain expertise. This presentation discusses three complementary approaches to make these decisions autonomously: active learning for experimental design, model predictive control, and agentic AI. These approaches are now usable for production science, not just toy problems, thanks to advances in scientific infrastructure that provide secure interfaces to scientific facilities. These workflows require orchestrating a diverse set of computational resources, including at the edge, in the cloud, and high-performance computing facilities to drive data capture, analysis, simulations, and AI tools. Examples of autonomous decision-making for manufacturing, chemistry, and neutron diffraction experiments are presented to highlight current capabilities and to chart a course toward a future that empowers scientists to unlock a host of new discoveries.