Research map / Probabilistic and scientific modeling / Variational and generative sampling
Efficient Training of Energy-Based Models Using Jarzynski Equality
Davide Carbone, Mengjian Hua, Simon Coste, Eric Vanden-Eijnden
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 |
|---|---|---|---|
| add Jarzynski walker weights to correct sampling bias | ULA-based EBM training · Gaussian mixture distributions | relative mode-weight recovery | Improved |
| use PCD without Jarzynski reweighting | EBM training · Gaussian mixture distributions | relative mode-weight recovery | Hurt |
| use CD with data reinitialization | EBM training · Gaussian mixture distributions | relative mode-weight recovery | Hurt |
| add Jarzynski reweighting | EBM training · MNIST with digits 2, 3, and 6 | relative mode-weight recovery | Improved |
| add systematic resampling rather than multinomial resampling (ablation) | Jarzynski-corrected EBM training · CIFAR-10 | FID: 32.18 -> 30.24 | Improved |
| add systematic resampling rather than multinomial resampling (ablation) | Jarzynski-corrected EBM training · CIFAR-10 | Inception Score (IS): 6.88 -> 6.97 | Improved |
| add data augmentation | PCD · CIFAR-10 | FID: 38.25 -> 36.43 | Improved |
| add data augmentation | PCD · CIFAR-10 | Inception Score (IS): 5.96 -> 6.54 | Improved |
| use Jarzynski-corrected training rather than PCD with data augmentation | EBM training · CIFAR-10 | FID: 36.43 -> 30.24 | Improved |
| use Jarzynski-corrected training rather than PCD with data augmentation | EBM training · CIFAR-10 | Inception Score (IS): 6.54 -> 6.97 | Improved |
| omit the resampling step (ablation) | Jarzynski-corrected EBM training · Gaussian mixture distributions | effective sample size | Hurt |
| resample too infrequently (ablation) | Jarzynski-corrected EBM training · Gaussian mixture distributions | relative mode-weight recovery | Hurt |
Closest papers by meaning
- BoltzNCE: Learning likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation (NeurIPS 2025 · similarity 0.87)
- Cascade of phase transitions in the training of energy-based models (NeurIPS 2024 · similarity 0.86)
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.86)
- Spherical Boltzmann machines: a solvable theory of learning and generation in energy-based models (NeurIPS 2026 · similarity 0.86)
- From Biased to Unbiased Dynamics: An Infinitesimal Generator Approach (NeurIPS 2024 · similarity 0.86)
- Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces (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.
