Mixed integer neural inverse design

In computational design and fabrication, neural networks are becoming important surrogates for bulky forward simulations. A long-standing, intertwined question is that of inverse design: how to compute a satisfies desired target performance? Here, we show the piecewise linear property, very common in everyday networks, allows an formulation based on mixed-integer programming. Our uncovers globally optimal or near solutions principled manner. Furthermore, our method significantly facilitates emerging, but challenging, combinatorial tasks, such as material selection. For problems where finding solution intractable, develop efficient yet near-optimal hybrid approach. Eventually, able find provably robust possible fabrication perturbations among multiple designs with similar performances. code data available at https://gitlab.mpi-klsb.mpg.de/nansari/mixed-integer-neural-inverse-design.