AGAN: Towards Automated Design of Generative Adversarial Networks
This addresses the need for automated, efficient GAN design, reducing reliance on human expertise and trial-and-error, though it is incremental as it builds on existing NAS and GAN concepts.
The paper tackles the problem of designing GAN architectures manually by introducing AGAN, a neural architecture search algorithm for GANs, which finds architectures that outperform state-of-the-art models on CIFAR-10 and achieve competitive performance on supervised tasks at 32x32 resolution.
Recent progress in Generative Adversarial Networks (GANs) has shown promising signs of improving GAN training via architectural change. Despite some early success, at present the design of GAN architectures requires human expertise, laborious trial-and-error testings, and often draws inspiration from its image classification counterpart. In the current paper, we present the first neural architecture search algorithm, automated neural architecture search for deep generative models, or AGAN for abbreviation, that is specifically suited for GAN training. For unsupervised image generation tasks on CIFAR-10, our algorithm finds architecture that outperforms state-of-the-art models under same regularization techniques. For supervised tasks, the automatically searched architectures also achieve highly competitive performance, outperforming best human-invented architectures at resolution $32\times32$. Moreover, we empirically demonstrate that the modules learned by AGAN are transferable to other image generation tasks such as STL-10.