image: Fig. 1 Contents of this review. When AI and structured light meet, the intersection facilitates the generation of new forms of light, the unravelling of structured light, and the application of structured light in diverse areas. Structured light too can be an intelligent machine: when passed through complex media, it acts as an optical neural network, offering new forms of AI. This self-referential cyclic interaction, a rapidly developing field of research, is rich with potential.
Credit: Zilong Zhang et al.
An exploding research topic of late is to think about light more deeply, giving it “structure” by controlling its many degrees of freedom. For instance, mixing polarization and spatial patterns results in vectorial light, allowing for tighter focusing in microscopy and cleaner machining in laser processing of materials. With a little more care to the type of pattern used, the light can be topological, giving it natural robustness to noisy systems. In optical communications, the many forms of structure light hold enormous potential for a new encoding alphabet, far exceeding what is possible with traditional approaches. To harness this enormous potential is challenging, requiring “intelligence” in the design, creation, deployment and detection steps. Here artificial intelligence (AI) based approaches have proven invaluable, allowing structured light to be unleashed in a myriad of applications, from classical to quantum, from microscopy to computing.
In a new paper published in eLight, a team of scientists, led by Professors Zilong Zhang from the School of Optics and Photonics, Beijing Institute of Technology, Andrew Forbes from University of the Witwatersrand in South Africa, Yijie Shen from Nanyang Technological University in Singapore, and co-workers, have revealed the exciting possibilities at the interface of structured light and AI, when structured light meets machine intelligence. They show that while structured light holds much potential, this richly textured light comes with deeply embedded complexity, making the design, analysis, and recognition of such complex light patterns highly non-trivial. The use of AI has come to the fore, offering innovative approaches and tools beyond the purely optical domain, not only for the design, characterization, and optimization of structured light but also for increasingly important roles in adding new functionalities and breaking old paradigms. For instance, it can remove unwanted complexity in optical communication, making noisy channels appear noise-free, and can be used to improve resolution in imaging and microscopy by taking into account both the light and the sample, both of which could be highly complex. In the quantum realm, AI can speed up the time to process unknown states and even design new experiments for novel forms of complex quantum states of light.
An exciting twist is the flip side of the coin, where complex light in complex media acts as a light-speed neural network. Here, the many patterns of structured light are deliberately chosen to mix in a complex way, one pattern mapping to many, a situation that looks just like a neural network. This opens the possibility for using the distortion of structured light – its unwanted complexity – for computing: optical-based “machines” for intelligence and learning at the speed of light.
Journal
eLight
Article Title
Structured light meets machine intelligence