A diffractive network reads the topology of structured light

A paper published on October 7 uses a diffractive optical neural network to identify skyrmion topologies. Two processing channels map input states into distinct output locations, allowing the system to classify the topological information directly.

The experiment tests 81 topologies and includes transmission of an image using a 15-level alphabet. Training in a reduced basis cuts the number of adjustable parameters compared with a pixel-based representation.

The useful idea is task-specific optical processing: extract the property the application needs rather than reconstruct everything and analyze it digitally. The demonstrated communication experiment remains narrower than a general-purpose optical computer.