Research map / Image generation / GANs
ExGAN: Adversarial Generation of Extreme Samples
Siddharth Bhatia, Arjit Jain, Bryan Hooi
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 |
|---|---|---|---|
| Use distribution shifting to focus training on increasingly extreme rainfall patterns (ablation) | ExGAN · US precipitation test extremes | FID: 0.0493 ± 0.0097 -> 0.0236 ± 0.0037 | Improved |
| Use distribution shifting to focus training on increasingly extreme rainfall patterns (ablation) | ExGAN · US precipitation test extremes | Reconstruction Loss: 0.0166 compared to 0.0172 | No clear effect |
| Replace DCGAN rejection sampling with EVT-based conditional generation and distribution shifting | US precipitation sampling at τ=0.01 · US precipitation, τ=0.01; 100 samples | sampling time: 7.564s -> 0.002s | Improved |
| Use ExGAN instead of DCGAN rejection sampling | US precipitation sampling at τ=0.05 · US precipitation, τ=0.05; 100 samples | sampling time: 1.230s -> 0.002s | Improved |
| Use ExGAN instead of DCGAN | US precipitation test extremes · US precipitation test extremes | FID: 0.0406 ± 0.0063 -> 0.0236 ± 0.0037 | Improved |
| Use ExGAN instead of DCGAN | US precipitation test extremes · US precipitation test extremes | Reconstruction Loss: 0.0292 -> 0.0172 | Improved |
| Use a more gradual distribution shift with c=0.90, k=27 instead of c=0.75, k=10 (ablation) | ExGAN · US precipitation test extremes | FID: 0.0236 ± 0.0037 -> 0.0223 ± 0.0121 | Improved |
| Use a more gradual distribution shift with c=0.90, k=27 instead of c=0.75, k=10 (ablation) | ExGAN · US precipitation test extremes | Reconstruction Loss: 0.0172 -> 0.0169 | Improved |
| Use c=0.90, k=27 instead of c=0.75, k=10 for distribution shifting (ablation) | ExGAN · US precipitation test extremes | MAPE: 3.14 ± 3.08 -> 3.05 ± 3.14 | Improved |
| Use c=0.24, k=2 instead of c=0.75, k=10 for distribution shifting (ablation) | ExGAN · US precipitation test extremes | FID: 0.0236 ± 0.0037 -> 0.0367 ± 0.0096 | Hurt |
| Use c=0.49, k=4 instead of c=0.75, k=10 for distribution shifting (ablation) | ExGAN · US precipitation test extremes | FID: 0.0236 ± 0.0037 -> 0.0304 ± 0.0109 | Hurt |
Closest papers by meaning
- EV-GAN: Simulation of extreme events with ReLU neural networks (JMLR 2022 · similarity 0.90)
- Pareto GAN: Extending the Representational Power of GANs to Heavy-Tailed Distributions (ICML 2021 · similarity 0.87)
- RareGAN: Generating Samples for Rare Classes (AAAI 2022 · similarity 0.86)
- Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models (AAAI 2018 · similarity 0.85)
- AlphaGAN: Fully Differentiable Architecture Search for Generative Adversarial Networks (TPAMI 2021 · similarity 0.85)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.85)
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.
