Optical multi-task learning using multi-wavelength diffractive deep neural networks

Abstract Photonic neural networks are brain-inspired information processing technology using photons instead of electrons to perform artificial intelligence (AI) tasks. However, existing architectures designed for a single task but fail multiplex different tasks in parallel within monolithic system due the competition that deteriorates model performance. This paper proposes novel optical multitask learning by designing multiwavelength diffractive deep (D 2 NNs) with joint optimization method. By encoding inputs into channels, can increase computing throughput and significantly alleviate multiple high accuracy. We design two-task four-task D NNs two four spectral respectively, classifying from MNIST, FMNIST, KMNIST, EMNIST databases. The numerical evaluations demonstrate that, under same network size, achieve higher classification accuracies than single-wavelength NNs. Furthermore, increasing simultaneously performing comparable respect individual training separately. Our work paves way developing wavelength-division multiplexing high-throughput neuromorphic photonic more general AI systems parallel.