Research map / Probabilistic and scientific modeling / Variational and generative sampling
Energy Discrepancies: A Score-Independent Loss for Energy-Based Models
Tobias Schröder, Zijing Ou, Jen Ning Lim, Yingzhen Li, Sebastian Josef Vollmer, Andrew Duncan
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 |
|---|---|---|---|
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · SVHN | FID: 29.44 -> 28.10 | Improved |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · CelebA 64x64 | FID: 37.87 -> 36.73 | Improved |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · CIFAR-10 | FID: 70.15 -> 73.58 | Hurt |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · CIFAR-10 | MSE: 0.020 -> 0.023 | Hurt |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · SVHN | MSE: 0.008 -> 0.006 | Improved |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · CelebA 64x64 | MSE: 0.013 -> 0.009 | Improved |
| train latent EBM with energy discrepancy instead of score matching | ED-LEBM · SVHN | FID: 34.44 -> 28.10 | Improved |
| train latent EBM with energy discrepancy instead of score matching | ED-LEBM · CIFAR-10 | FID: 77.82 -> 73.58 | Improved |
| train latent EBM with energy discrepancy instead of score matching | ED-LEBM · CelebA 64x64 | FID: 41.21 -> 36.73 | Improved |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · MNIST anomaly, heldout digit 4 | AUPRC: 0.630 -> 0.740 | Improved |
| train latent EBM with energy discrepancy instead of contrastive divergence | ED-LEBM · MNIST anomaly, heldout digit 1 | AUPRC: 0.336 -> 0.342 | Improved |
| train latent EBM with energy discrepancy instead of score matching | ED-LEBM · MNIST anomaly, heldout digit 1 | AUPRC: 0.285 -> 0.342 | Improved |
Closest papers by meaning
- Learning normalized image densities via dual score matching (NeurIPS 2025 · similarity 0.88)
- Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces (NeurIPS 2024 · similarity 0.87)
- Energy-Tweedie: Score meets Score, Energy meets Energy (NeurIPS 2026 · similarity 0.87)
- A Theoretical Framework For Overfitting In Energy-based Modeling (ICML 2025 · similarity 0.86)
- Bridging Mini-Batch and Asymptotic Analysis in Contrastive Learning: From InfoNCE to Kernel-Based Losses (ICML 2024 · similarity 0.85)
- Learning Energy-Based Models from Stochastic Interpolants using Spatiotemporal Differences (NeurIPS 2026 · similarity 0.85)
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.
