Research map / Probabilistic and scientific modeling / Variational and generative sampling
Divergence Triangle for Joint Training of Generator Model, Energy-Based Model, and Inferential Model
Check 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 |
|---|---|---|---|
| Jointly train generator, energy-based model, and inference model with the divergence triangle objective rather than the strongest reported baseline | Image generation · CIFAR-10 | Inception Score: 7.07 -> 7.23 | Helped |
| Jointly train generator, energy-based model, and inference model with the divergence triangle objective rather than the strongest reported baseline | Image generation · CelebA | FID: 36.36 -> 31.92 | Helped |
| Use the divergence triangle model instead of the strongest reported baseline | Test image reconstruction · CIFAR-10 test set | MSE: 0.034 -> 0.028 | Helped |
| Use the divergence triangle model instead of the strongest reported baseline | Test image reconstruction · CelebA hold-out set | MSE: 0.039 -> 0.030 | Helped |
| Use the divergence triangle model instead of ABP | Incomplete-image recovery · CelebA P.5 occlusion | Recovery error: 0.0379 -> 0.0380 | Hurt |
| Use the divergence triangle model instead of ABP | Incomplete-image recovery · CelebA P.7 occlusion | Recovery error: 0.0428 -> 0.0430 | Hurt |
| Use the divergence triangle model instead of ABP | Incomplete-image recovery · CelebA MB10 occlusion | Recovery error: 0.1070 -> 0.1060 | Helped |
| Use the divergence triangle model instead of ABP | Incomplete-image recovery · CelebA B20 occlusion | Recovery error: 0.0633 -> 0.0621 | Helped |
| Use the divergence triangle model instead of ABP | Incomplete-image recovery · CelebA B30 occlusion | Recovery error: 0.0757 -> 0.0733 | Helped |
| Train with the divergence triangle rather than VAE | Incomplete-image recovery · CelebA occlusion patterns | Recovery error | Helped |
Closest papers by meaning
- Improved Contrastive Divergence Training of Energy-Based Models (ICML 2021 · similarity 0.87)
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.87)
- Joint Training of Variational Auto-Encoder and Latent Energy-Based Model (CVPR 2020 · similarity 0.87)
- Training Deep Energy-Based Models with f-Divergence Minimization (ICML 2020 · similarity 0.86)
- Scalable Energy-Based Models via Adversarial Training: Unifying Discrimination and Generation (ICLR 2026 · similarity 0.86)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.83)
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.