Research map / Image generation / GANs
Generalized Adversarially Learned Inference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace ALI's two-way matching with GALI-4 multi-distribution feedback | ALI · CelebA test | pixel-MSE: 0.074 ± 0.004 -> 0.036 ± 0.002 | Helped |
| Replace ALI's two-way matching with GALI-4 multi-distribution feedback | ALI · CelebA | FID: 24.49 -> 23.11 | Helped |
| Replace ALI's two-way matching with GALI-4 multi-distribution feedback | ALI · SVHN test, linear SVM with 1000 labeled examples | misclassification rate (%): 19.05 ± 0.53 -> 16.58 ± 0.38 | Helped |
| Add 4 tuple classes to use 8 classes (ablation) | GALI-4 · SVHN test | feature-MSE: 0.0717 ± 0.0067 -> 0.066 ± 0.0060 | Helped |
| Add 4 tuple classes to use 8 classes (ablation) | GALI-4 · SVHN test, linear SVM with 1000 labeled examples | misclassification rate (%): 16.58 ± 0.38 -> 15.82 ± 0.43 | Helped |
| Add pretrained-model features as learned-knowledge feedback (ablation) | GALI-4 · CelebA test | feature-MSE: 0.201 ± 0.012 -> 0.131 ± 0.008 | Helped |
| Add pretrained-model features as learned-knowledge feedback (ablation) | GALI-4 · CelebA | FID: 23.11 -> 10.13 | Helped |
| Use an L2 feature reconstruction loss instead of joint-distribution learned-knowledge feedback (ablation) | GALI-PT · CelebA reconstruction | pixel-MSE: 0.032 ± 0.0019 -> 0.053 ± 0.003 | Hurt |
| Add pretrained-model feature L2 regression to ALI (ablation) | ALI · CelebA reconstruction | pixel-MSE: 0.074 ± 0.004 -> 0.068 ± 0.0053 | Helped |
| Add mixed-image self-supervision (ablation) | GALI-4 · CelebA inpainting | pixel-MSE: 0.046 ± 0.0066 -> 0.031 ± 0.0050 | Helped |
| Add mixed-image self-supervision (ablation) | GALI-4 · CelebA inpainting | feature-MSE: 0.285 ± 0.012 -> 0.164 ± 0.010 | Helped |
| Add pixel-level L2 reconstruction error (ablation) | ALI · CelebA | FID: 24.49 -> 36.91 | Hurt |
Closest papers by meaning
- Adversarially Learned Inference (ICLR 2017 · similarity 0.91)
- Optimizing the Latent Space of Generative Networks (ICML 2018 · similarity 0.90)
- SGAN: An Alternative Training of Generative Adversarial Networks (CVPR 2018 · similarity 0.90)
- GLeaD: Improving GANs With a Generator-Leading Task (CVPR 2023 · similarity 0.90)
- TRAINING GENERATIVE ADVERSARIAL NETWORKS VIA PRIMAL-DUAL SUBGRADIENT METHODS: A LAGRANGIAN PERSPECTIVE ON GAN (ICLR 2018 · similarity 0.89)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.89)
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.