Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Energy-based Model via Dual-MCMC Teaching
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 dual-MCMC teaching with a generator and inference model to initialize MCMC | Energy-based image modeling · CIFAR-10 | FID: 30.10 -> 9.26 | Helped |
| Use dual-MCMC teaching rather than Divergence Triangle | Energy-based image modeling · CelebA-64 | FID: 18.21 -> 5.15 | Helped |
| Initialize latent posterior MCMC with the inference model rather than noise (ablation) | Generator posterior inference · CIFAR-10 | MSE: 0.0214 -> 0.0072 | Helped |
| Initialize latent posterior MCMC with the inference model rather than noise (ablation) | Generator posterior inference · CelebA-64 | MSE: 0.0227 -> 0.0164 | Helped |
| Increase latent-space Langevin steps from 10 to 30 without inference initialization (ablation) | Dual-MCMC teaching · CIFAR-10 | FID: 17.32 -> 14.51 | Helped |
| Increase inference-initialized latent-space Langevin steps from 10 to 30 (ablation) | Dual-MCMC teaching · CIFAR-10 | FID: 9.26 -> 9.18 | No clear effect |
| Increase inference-initialized latent-space Langevin steps from 10 to 30 (ablation) | Dual-MCMC teaching · CIFAR-10 | Time (s): 1.594 -> 2.112 | Hurt |
| Increase EBM-guided data-space Langevin steps from 10 to 30 (ablation) | EBM sampling · CIFAR-10 | FID: 14.78 -> 9.26 | Helped |
| Increase EBM-guided data-space Langevin steps from 30 to 50 (ablation) | EBM sampling · CIFAR-10 | FID: 9.26 -> 9.07 | No clear effect |
| Increase hidden features from nef=256 to nef=512 (ablation) | Dual-MCMC teaching model · CIFAR-10 | FID: 11.19 -> 9.26 | Helped |
| Increase hidden features from nef=512 to nef=1024 (ablation) | Dual-MCMC teaching model · CIFAR-10 | FID: 9.26 -> 8.45 | Helped |
| Use latent posterior MCMC revision before generation (ablation) | Image inpainting · CelebA-64, M20 | PSNR: 21.035 -> 24.976 | Helped |
Closest papers by meaning
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.88)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.87)
- Learning Latent Space Hierarchical EBM Diffusion Models (ICML 2024 · similarity 0.87)
- BoltzNCE: Learning likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation (NeurIPS 2025 · similarity 0.87)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 2025 · similarity 0.86)
- From Biased to Unbiased Dynamics: An Infinitesimal Generator Approach (NeurIPS 2024 · 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.