Research map / Image generation / GANs
AlphaGAN: Fully Differentiable Architecture Search for Generative Adversarial Networks
Yuesong Tian, Li Shen, Li Shen, Guinan Su, Zhifeng Li, Wei Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| search the discriminator architecture too (ablation) | alphaGAN(l) · CIFAR-10 | FID: 11.34 -> 18.07 | Hurt |
| search the discriminator architecture too (ablation) | alphaGAN(s) · CIFAR-10 | FID: 10.35 -> 15.56 | Hurt |
| update weights w_G only to obtain G-bar instead of also alpha_G (ablation) | alphaGAN(s) · CIFAR-10 | FID: 18.85 -> 10.35 | Improved |
| update alpha_G only to obtain G-bar (ablation) | alphaGAN(l) · CIFAR-10 | FID: 11.34 -> 43.99 | Hurt |
| use bi-level instead of single-level optimization (ablation) | alphaGAN(s) · CIFAR-10 | FID: 23.40 -> 10.35 | Improved |
| add Gumbel-max trick (ablation) | alphaGAN(l) · CIFAR-10 | FID: 11.38 -> 20.69 | Hurt |
| add warm-up protocol (ablation) | alphaGAN(l) · CIFAR-10 | FID: 11.38 -> 15.49 | Hurt |
| use smaller step sizes (alphaGAN(s) vs (l)) to cut search cost | alphaGAN · CIFAR-10 | search cost GPU-hours: 22 -> 3 | Improved |
| search dynamic latent W_i in micro space (ablation) | alphaGAN(s) on StyleGAN2 · CelebA | FID: 2.07 -> 1.94 | Improved |
| replace alphaGAN with AdversarialNAS (micro + dynamic W) | StyleGAN2 search · LSUN-church | FID: 2.86 -> 4.07 | Hurt |
| apply alphaGAN to BigGAN | BigGAN · CIFAR-10 | FID: 14.73 -> 10.27 | Improved |
| remove deconvolution from search space or make interpolation learnable | alphaGAN(s) · CIFAR-10 | failure cases: 45.88 -> 15.12 | Improved |
Closest papers by meaning
- EAGAN: Efficient Two-Stage Evolutionary Architecture Search for GANs (ECCV 2022 · similarity 0.92)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.91)
- AdversarialNAS: Adversarial Neural Architecture Search for GANs (CVPR 2020 · similarity 0.91)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.91)
- DO-GAN: A Double Oracle Framework for Generative Adversarial Networks (CVPR 2022 · similarity 0.90)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
