Optoelectronic generative adversarial networks
Jumin Qiu, Ganqing Lu, Tingting Liu, Dejian Zhang, Shuyuan Xiao, Tianbao Yu · Communications Physics · 2025
Recent breakthroughs in artificial intelligence generative content technology are driving transformational change. Diffractive optical networks offer a promising solution for high-speed, low-power generative models. However, the generative capabilities of optical computing remain underexplored. Here, we present the implementation of a generative model on an optoelectronic computing architecture based on generative adversarial networks, termed the optoelectronic generative adversarial network. The network strategically distributes the generator and discriminator across optical and electronic components, which are seamlessly integrated to leverage the unique strengths of each computing paradigm and take advantage of transfer learning. The network can efficiently and rapidly process complex generative tasks. The performance of the network is demonstrated through three generative tasks: image generation, conditional generation, and image restoration. By integrating the advantages of optical and electronic computing, the network advances the development of more powerful and accessible generative models, unlocking new creative possibilities across diverse applications. Recent advances in artificial intelligence require massive resources, and optical computing is emerging as a fast and energy-efficient alternative for AI tasks. In this work, the authors develop an optoelectronic generative adversarial network that combines optical and electronic components to achieve fast, energy-efficient image generation.