Figure 5
(a) Example of vertical and (b) horizontal reorientation of inverse ML output and subsequent generation of randomly distributed solution points about the adjusted geometric solution. Example shown is using an aspect ratio limitation of 5. These solutions are evaluated using both FDTD and the surrogate, the most optimal new solution is using a process described in the methods section. (c) The identified optimal point at each aspect ratio is shown for each aspect ratio. While the solution is not as optimal as the original ML generated geometry, the we can still identify geometric designs with exceptional performance.
