Chaotic optical combs supply randomness for a photonic neural network

An Optics Letters paper in the October 15 issue uses chaotic Kerr optical frequency combs as a source of randomness for probabilistic photonic computing. The experimental link combines optical modulators and a microring array with a hybrid Bayesian neural network.

The researchers test blood-cell images and detection of examples outside the training distribution. That second task matters because a classifier can be confidently wrong when it encounters unfamiliar inputs.

The result concerns a specific experimental system and workload. It does not establish clinical diagnostic performance or a general replacement for electronic AI accelerators, but it explores a useful role for randomness that optical hardware naturally produces.