Learning-based Optimization for Complex Physical Systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Learning-based optimization (i.e., reinforcement learning) is known for its strength in generalizability and scalability. Generalizability indicates the knowledge learned in optimizing the training data can be directly applied to optimize the unseen testing scenarios without retraining. Scalability represents the capability of handling giant design space. In this talk, we will discuss the recent research work on developing learning-based optimization for complex physical systems. We select two cases as examples. One is the fixture layout design for reducing shape deformation of large-scale sheet parts in the assembly process. The other is optimizing the control parameters in the Brookhaven Alternating Gradient Synchrotron (AGS) to improve intensity and emittance preservation after the RF bunch merge.