Photonic quantum generative adversarial networks for classical data

When Generative Adversarial Networks (GANs) first emerged, they marked a breakthrough in the field of classical machine learning. Researchers have since designed quantum versions of the algorithm, both for the generation of classical and quantum data, but most work so far has focused on qubit-based architectures. In this article, we focus on photonic quantum computing and present a quantum GAN based on linear optical circuits and Fock-space encoding for the generation of classical data. We explore the trainability and the performance of the model in a proof-of-concept image generation scenario. We then conduct an experiment where we train our quantum GAN on Quandela's photonic quantum processor Ascella.