IFIMAC researchers have shown how generative Artificial Intelligence (AI) can make the design of photonic nanostructures faster and more efficient. Generative AI technologies are revolutionizing the way to create texts and images, as evident by the rapid advances of new technologies such as ChatGTP, Midjourney, DALL.E and Stable Diffusion, to name a few. Most importantly, rather than providing a single solution, Generative AI technologies can provide full distributions of possible solutions.
In the work published in Machine Learning: Science and Technology by Michel Frising, Jorge Bravo-Abad and Ferry Prins demonstrate the superior performance of generative AI when dealing with complex parameter spaces. Most traditional optimization routines assume an invertible one-to-one mapping between the design parameters and the target performance. However, comparable or even identical performance may be realized by different designs, yielding a multimodal distribution of possible solutions to the inverse design problem which confuses the optimization algorithm. Here, Frising et al. show how a generative modeling approach based on invertible neural networks can provide the full distribution of possible solutions to the inverse design problem and resolve the ambiguity of nanodevice inverse design problems featuring multimodal distributions. The authors implement a Conditional Invertible Neural Network (cINN) and apply it to a proof-of-principle nanophotonic problem, consisting in tailoring the transmission spectrum of a metallic film milled by subwavelength indentations. They compare their approach with the commonly used conditional Variational Autoencoder (cVAE) framework and show the superior flexibility and accuracy of the proposed cINNs when dealing with multimodal device distributions. The work shows that invertible neural networks provide a valuable and versatile toolkit for advancing inverse design in nanoscience and nanotechnology. [Full article]
