Research map / Probabilistic and scientific modeling / Variational and generative sampling
Generalized Energy Based Models
Michael Arbel, Liang Zhou, Arthur Gretton
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 the learned energy to sample from the latent posterior instead of sampling from the base alone (ablation) | GEBM with SNGAN ConvNet · Cifar10 | FID: 32.03 -> 23.02 | Improved |
| Use the learned energy to sample from the latent posterior instead of sampling from the base alone (ablation) | GEBM with SNGAN ConvNet · LSUN | FID: 21.67 -> 15.25 | Improved |
| Use the learned energy to sample from the latent posterior instead of sampling from the base alone (ablation) | GEBM with SNGAN ConvNet · CelebA | FID: 6.91 -> 5.21 | Improved |
| Use the learned energy to sample from the latent posterior instead of sampling from the base alone (ablation) | GEBM with SNGAN ConvNet · ImageNet | FID: 19.37 -> 13.94 | Improved |
| Use KLA rather than ULA for latent-posterior sampling (ablation) | GEBM with SNGAN ConvNet · Cifar10 | FID: 23.02 -> 24.29 | Hurt |
| Use KLA rather than ULA for latent-posterior sampling (ablation) | GEBM with SNGAN ConvNet · LSUN | FID: 16.23 -> 15.25 | Improved |
| Train with KALE (F) rather than direct maximum likelihood | Real NVP density model · RedWine, Whitewine, Parkinsons, Hepmass, Miniboone | NLL: 11.98 -> 11.19; 13.05 -> 12.66; 14.5 -> 13.26; 24.89 -> 24.66; 42.28 -> 38.35 | Mixed |
| Train with KALE (F) rather than contrastive divergence | Real NVP density model · RedWine, Whitewine, Parkinsons, Hepmass, Miniboone | NLL: 11.88 -> 11.19; 13.01 -> 12.66; 14.06 -> 13.26; 22.89 -> 24.66; 39.36 -> 38.35 | Mixed |
| Use KALE (F) rather than KALE (DV) | Real NVP density model · RedWine, Whitewine, Parkinsons, Hepmass, Miniboone | NLL: 11.6 -> 11.19; 12.77 -> 12.66; 13.26 -> 13.26; 26.56 -> 24.66; 46.48 -> 38.35 | Mixed |
| Rescale the energy by beta=100 for colder sampling | GEBM · Image-generation tasks | FID | Improved |
Closest papers by meaning
- Implicit Generation and Modeling with Energy Based Models (NeurIPS 2019 · similarity 0.89)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.88)
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.88)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.86)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.86)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 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.
