Research map / Architectures and efficiency / Neural architecture search
AutoGAN: Neural Architecture Search for Generative Adversarial Networks
Xinyu Gong, Shiyu Chang, Yifan Jiang, Zhangyang Wang
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 |
|---|---|---|---|
| add parameter dynamic-resetting to the shared-GAN search (ablation) | AutoGAN search · CIFAR-10 | search time: 72 hours -> 43 hours | Improved |
| add parameter dynamic-resetting to the shared-GAN search (ablation) | AutoGAN search · CIFAR-10 | Inception score | No clear effect |
| use multi-level architecture search instead of single-level search (ablation) | AutoGAN · CIFAR-10 | Inception score | Improved |
| use reciprocal FID as the controller reward instead of IS (ablation) | AutoGAN · CIFAR-10 | IS and FID | No clear effect |
| use RL controller search instead of random search with weight sharing (ablation) | AutoGAN · CIFAR-10 | Inception score: 8.09 -> 8.55 | Improved |
| use RL controller search instead of random search with weight sharing (ablation) | AutoGAN · CIFAR-10 | FID: 17.34 -> 12.42 | Improved |
| use RL controller search instead of random search without weight sharing (ablation) | AutoGAN · CIFAR-10 | FID: 21.39 -> 12.42 | Improved |
| use RL controller search instead of random search without weight sharing (ablation) | AutoGAN · CIFAR-10 | Inception score: 7.97 -> 8.55 | Improved |
| search the generator architecture with the same building blocks instead of hand-designing it | SN-GAN · CIFAR-10 | FID: 21.7 -> 12.42 | Improved |
| search the generator architecture with the same building blocks instead of hand-designing it | SN-GAN · CIFAR-10 | Inception score: 8.22 -> 8.55 | Improved |
| transfer the CIFAR-10 searched architecture to STL-10 without re-searching | AutoGAN · STL-10 | FID: 40.1 -> 31.01 | Improved |
| transfer the CIFAR-10 searched architecture to STL-10 without re-searching | AutoGAN · STL-10 | Inception score: 9.34 -> 9.16 | Hurt |
Later papers that compared with this method
3 later accepted papers report AutoGAN as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Auto-GAS: Automated Proxy Discovery for Training-free Generative Architecture Search (ECCV 2024 · better on 1 of 1 shared result)
- Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS 2021 · better on 1 of 1 shared result)
- AdversarialNAS: Adversarial Neural Architecture Search for GANs (CVPR 2020 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EAGAN: Efficient Two-Stage Evolutionary Architecture Search for GANs (ECCV 2022 · similarity 0.93)
- AdversarialNAS: Adversarial Neural Architecture Search for GANs (CVPR 2020 · similarity 0.92)
- AlphaGAN: Fully Differentiable Architecture Search for Generative Adversarial Networks (TPAMI 2021 · similarity 0.90)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 2024 · similarity 0.89)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.87)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.86)
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.
