Mixed Integer Neural Inverse Design
Event Type
Technical Paper
Interest Areas
Research & Education
Presentation Types
Registration Categories
Full Conference Supporter
Full Conference
Virtual Conference Supporter
Virtual Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
DescriptionNeural networks are among the most common modeling surrogates, and their inversion is crucial to finding designs with desirable performances. We have reformulated and solved the neural inverse design problem as a mixed integer linear programming, which is capable of finding globally optimum designs and handling combinatorial design constraints.