Research map / Probabilistic and scientific modeling / Variational and generative sampling
Training Deep Energy-Based Models with f-Divergence Minimization
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 f-EBM with KL instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | Inception: 8.30 -> 8.11 $\pm$ .06 | Hurt |
| Use f-EBM with KL instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | FID: 37.90 -> 37.36 | Helped |
| Train with Reverse KL instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | Inception: 8.30 -> 8.49 $\pm$ .09 | Helped |
| Train with Reverse KL instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | FID: 37.90 -> 33.25 | Helped |
| Train with Squared Hellinger instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | Inception: 8.30 -> 8.57 $\pm$ .08 | Helped |
| Train with Squared Hellinger instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | FID: 37.90 -> 32.19 | Helped |
| Train with Jensen-Shannon instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | Inception: 8.30 -> \textbf{8.61} $\pm$ .06 | Helped |
| Train with Jensen-Shannon instead of Contrastive Divergence (KL) | EBM · CIFAR-10 conditional generation | FID: 37.90 -> \textbf{30.86} | Helped |
| Use f-EBM with KL instead of Contrastive Divergence (KL) | EBM · CelebA (32 × 32) unconditional generation | FID: 35.97 -> 37.41 | Hurt |
| Train with Reverse KL instead of Contrastive Divergence (KL) | EBM · CelebA (32 × 32) unconditional generation | FID: 35.97 -> 26.75 | Helped |
| Train with Jensen-Shannon instead of Contrastive Divergence (KL) | EBM · CelebA (32 × 32) unconditional generation | FID: 35.97 -> 26.53 | Helped |
| Train with Squared Hellinger instead of Contrastive Divergence (KL) | EBM · CelebA (32 × 32) unconditional generation | FID: 35.97 -> \textbf{23.67} | Helped |
Closest papers by meaning
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.91)
- Guiding Energy-based Models via Contrastive Latent Variables (ICLR 2023 · similarity 0.88)
- Improved Contrastive Divergence Training of Energy-Based Models (ICML 2021 · similarity 0.88)
- Learning Energy-Based Model with Variational Auto-Encoder as Amortized Sampler (AAAI 2021 · similarity 0.88)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.87)
- Joint Learning of Energy-based Models and their Partition Function (ICML 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.