Research map / Optimization / Optimizers
Multi-objective training of Generative Adversarial Networks with multiple discriminators
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 |
|---|---|---|---|
| Use hypervolume maximization instead of average loss for the generator | Multiple-discriminator GAN on MNIST · MNIST | FID | Helped |
| Use MGD instead of hypervolume maximization | Multiple-discriminator GAN · MNIST | FID | Helped |
| Use MGD instead of hypervolume maximization | Multiple-discriminator GAN · MNIST | training time: approximately 7 times slower | Hurt |
| Increase discriminator count from 8 to 16 (ablation) | Hypervolume GAN · Stacked MNIST (10k) | Modes (Max 1000): 679.2 ± 5.9 -> 998.0 ± 1.8 | Helped |
| Increase discriminator count from 8 to 16 (ablation) | Hypervolume GAN · Stacked MNIST (10k) | KL: 1.139 ± 0.011 -> 0.120 ± 0.004 | Helped |
| Increase discriminator count from 16 to 24 (ablation) | Hypervolume GAN · Stacked MNIST (10k) | Modes (Max 1000): 998.0 ± 1.8 -> 998.3 ± 1.1 | No clear effect |
| Increase discriminator count from 16 to 24 (ablation) | Hypervolume GAN · Stacked MNIST (10k) | KL: 0.120 ± 0.004 -> 0.116 ± 0.003 | Helped |
| Increase discriminator count from 8 to 16 (ablation) | Hypervolume GAN · Stacked MNIST (26k) | Modes (Max 1000): 776.8 ± 6.4 -> 1000.0 ± 0.0 | Helped |
| Increase discriminator count from 8 to 24 (ablation) | Hypervolume GAN · CIFAR-10 | FID-ResNet: 4.22 -> 1.89 | Helped |
| Increase discriminator count from 1 to 24 (ablation) | LSGAN with hypervolume training · CIFAR-10 | FID-ResNet: 4.55 -> 1.91 | Helped |
| Increase discriminator count from 1 to 24 (ablation) | HingeGAN with hypervolume training · CIFAR-10 | FID-ResNet: 6.17 -> 2.25 | Helped |
| Increase discriminator count from 1 to 24 (ablation) | DCGAN with hypervolume training · CIFAR-10 | FLOPS: 8e10 -> 5e11 | Hurt |
Closest papers by meaning
- MCL-GAN: Generative Adversarial Networks with Multiple Specialized Discriminators (NeurIPS 2022 · similarity 0.90)
- Generative Multi-Adversarial Networks (ICLR 2017 · similarity 0.90)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.90)
- TRAINING GENERATIVE ADVERSARIAL NETWORKS VIA PRIMAL-DUAL SUBGRADIENT METHODS: A LAGRANGIAN PERSPECTIVE ON GAN (ICLR 2018 · similarity 0.90)
- Dualing GANs (NeurIPS 2017 · similarity 0.89)
- Dual Discriminator Generative Adversarial Nets (NeurIPS 2017 · 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.