Research map / Probabilistic and scientific modeling / Molecules and proteins
Versatile Energy-Based Probabilistic Models for High Energy Physics
Taoli Cheng, Aaron Courville
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 an EBM instead of a β-VAE for jet generation | QCD jet generator · QCD jet generation | JSD (M): 11.0 -> $3.2 \pm 2.0$ | Improved |
| Jointly train the generative EBM with a classifier | EBM anomaly detector · OOD H jet tagging | AUC: $0.794 \pm 0.032$ -> $0.815 \pm 0.018$ | Improved |
| Use the score-function norm instead of energy as the anomaly score | EBM · Top jet tagging | AUC: $0.679 \pm 0.009$ -> $0.628 \pm 0.017$ | Hurt |
| Use the score-function norm instead of energy as the anomaly score | EBM · OOD H jet tagging | AUC: $0.794 \pm 0.032$ -> $0.738 \pm 0.033$ | Hurt |
| Use the score-function norm instead of energy as the anomaly score | EBM-CLF · OOD jet tagging | AUC: $0.703 \pm 0.020$ -> $0.679 \pm 0.040$; $0.815 \pm 0.018$ -> $0.722 \pm 0.028$ | Mixed |
| Reduce the discriminative log-likelihood weight to κ=0.5 (ablation) | EBM-CLF · QCD/W/Top 3-way classification | Top-1 Accuracy: 0.850 -> 0.708 | Hurt |
| Reduce the discriminative log-likelihood weight to κ=0.1 (ablation) | EBM-CLF · QCD/W/Top 3-way classification | Top-1 Accuracy: 0.850 -> 0.679 | Hurt |
| Replace the transformer energy function with an MLP (ablation) | EBM jet generator · QCD jet generation | Generation quality | Hurt |
| Replace Langevin dynamics with Hamiltonian Monte Carlo (ablation) | EBM jet generator · QCD jet generation | JSD (p_T): 24 | Hurt |
| Replace Langevin dynamics with Hamiltonian Monte Carlo (ablation) | EBM jet generator · QCD jet generation | JSD (M): 30 | Hurt |
| Use fewer MCMC steps, 5 or 10 rather than the selected 24 (ablation) | EBM training · QCD jet generation | Training stability | Hurt |
| Increase training MCMC steps to 60 or longer rather than the selected 24 (ablation) | EBM training · QCD jet generation | Generation quality | No clear effect |
Closest papers by meaning
- End-To-End Latent Variational Diffusion Models for Inverse Problems in High Energy Physics (NeurIPS 2023 · similarity 0.85)
- Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modeling (NeurIPS 2025 · similarity 0.83)
- Patch Hierarchical Attention Transformers for Efficient Particle Jet Tagging (NeurIPS 2026 · similarity 0.82)
- Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physics (NeurIPS 2024 · similarity 0.81)
- Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces (NeurIPS 2024 · similarity 0.81)
- BoltzNCE: Learning likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation (NeurIPS 2025 · similarity 0.81)
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.
