DEGAS: differentiable efficient generator search

Sivan Doveh*, Raja Giryes

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

5 Scopus citations


Network architecture search achieves state-of-the-art results in various tasks such as classification and semantic segmentation. Recently, a reinforcement learning-based approach has been proposed for generative adversarial networks (GANs) search. In this work, we propose an alternative strategy for GAN search by using a proxy task instead of common GAN training. Our method is called differentiable efficient generator search, which focuses on efficiently finding the generator in the GAN. Our search algorithm is inspired by the differential architecture search strategy and the global latent optimization procedure. This leads to both an efficient and stable GAN search. After the generator architecture is found, it can be plugged into any existing framework for GAN training. For consistency-term GAN, which we use in this work, the new model outperforms the original inception score results by 0.25 for CIFAR-10.

Original languageEnglish
Pages (from-to)17173-17184
Number of pages12
JournalNeural Computing and Applications
Issue number24
StatePublished - Dec 2021


FundersFunder number


    • Computer vision
    • Generative adversarial networks
    • Network architecture search
    • Stable training


    Dive into the research topics of 'DEGAS: differentiable efficient generator search'. Together they form a unique fingerprint.

    Cite this