Proposal of coherent optical neural networks whose behavior is controlled by optical frequency

Complex-valued neural networks enable us to construct coherent-type optical neural networks where we can use sophisticated artificial neural mechanisms. On the basis of the complex-valued neural networks, the authors propose coherent optical neural networks whose behavior is controlled by the optical frequency. Photon energy is introduced in the learning process as a network parameter. The learning process is derived and demonstrated successfully in experiments. These results will be useful for ultimate optical neural networks in the future