An evolutionary quantum generative adversarial network

Generative adversarial networks have been widely used in image generation, text, and image enhancement. But classical generative adversarial training takes longer, and at the same time, there is a risk of mode collapse. A quantum version of generative adversarial networks has been proposed, and the general quantum version can greatly reduce the parameter set. But the existing quantum versions of generative adversarial targets are all single, so we propose a framework called evolutionary quantum generative adversarial networks to improve the performance of quantum generative adversarial networks. We couple them together by using quantum circuits as generators and classical neural networks as discriminators. And different adversarial training targets are used as mutation operations, and the fitness is calculated separately as an evaluation criterion for selecting the next generation. In this way, our model can overcome the limitations of adversarial singularity and can train faster and less expensively. Finally, we also validate our framework on two datasets by simulation.