Deep diffractive optical neural networks for detecting skyrmionic topologies of light and their live training.

Deep diffractive neural networks can reliably detect optical skyrmions and their topological invariant N with high accuracy, even under noisy conditions.

• Why it matters: Optical skyrmions encode information in topological states that are difficult to distinguish with traditional methods, limiting their practical application in optical communication and information processing.
• What they did: A deep diffractive neural network with two channels and five layers per channel was trained in real time, reducing training parameters by 1000-fold, to identify 81 topologies from vectorial Laguerre-Gaussian modes.
• The result: The detector accurately classifies topologies with minimal crosstalk, enabling robust topological encoding and decoding, paving the way for topology-based optical communication systems.