Time-series classification with an all-optical recurrent neuron
George Mourgias-Alexandris, Nikolaos Passalis, George Dabos, Angelina R. Totovic, Anastasios Tefas, Nikos Pleros · 2020
We present an all-optical recurrent neuron comprising a novel space rotator and a sigmoid activation unit and demonstrate experimentally time-series classification. Real-time recognition of both RZ and NRZ 3-bit-long bit sequences with 100psec optical pulses is successfully demonstrated, revealing an average accuracy of >91%.