Light carries information in its intensity, phase, spectrum, polarisation, space and time, yet a conventional photodetector produces only a single photocurrent. Computational light-field detection encodes this multidimensional information into compact measurements and reconstructs it with algorithms.

This review examines how artificial intelligence is changing both sides of the process, from designing nanophotonic front ends to recovering information from sparse measurements. The authors argue for co-designing hardware and algorithms, for example with differentiable digital twins, and note that real-world impact will depend on generalisation, physical consistency, interpretability and robustness. Zhipei Sun of QMAT's Photonics group at Aalto University co-authored the review with researchers at the University of Cambridge, Zhejiang University, Shanghai Jiao Tong University and other institutions.

Image: QMAT