Research map / Image generation / Fast diffusion sampling
RNE: plug-and-play diffusion inference-time control and energy-based training
Jiajun He, José Miguel Hernández-Lobato, Yuanqi Du, Francisco Vargas
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 |
|---|---|---|---|
| replace summed scores with RNC sampling and target process | Sum score · Multi-target SBDD | better than known: 0.345 -> 0.649 | Improved |
| use RNC with c_b=0.2 instead of FKC | FKC · Multi-target SBDD | better than known: 0.608 -> 0.649 | Improved |
| use RNC with c_a=1, c_b=0 instead of FKC | FKC · Multi-target SBDD | better than known: 0.608 -> 0.589 | No clear effect |
| replace summed scores with FKC/RNC SMC | Sum score · Multi-target SBDD | diversity: 0.881 -> 0.814 | Hurt |
| add RNC SMC to reward-gradient guidance | guidance without SMC · Maze stitching task 4 | success rate: 0.288 -> 1.000 | Improved |
| add RNE regularisation to thermodynamic integration | TI without RNE · ALDP solvation free energy | free energy estimate: 27.30 -> 29.28 | Improved |
| add RNE regularisation to denoising score matching | DSM · 2D GMM density | learned density | Improved |
| add RNE regularisation or dual score matching to DSM | DSM · 100D GMM | learned energy accuracy | Improved |
| choose c_a+c_b near 1 (within 1 +/- 0.2) in RNC (ablation) | RNC · ALDP annealing | importance weight variance: c_a+c_b within 1 +/- 0.2 | Improved |
| use general (c_a,c_b) instead of c_a=1, c_b=0 (ablation) | RNC · ALDP annealing | W2 and TVD | Improved |
| increase batch size from 100 to 5000 particles with RNC instead of FKC (ablation) | FKC · ALDP annealing | W2 distance | Improved |
| add RNE regularisation to DSM | DSM · ALDP diffusion quality | sample quality | No clear effect |
Closest papers by meaning
- Reverse Transition Kernel: A Flexible Framework to Accelerate Diffusion Inference (NeurIPS 2024 · similarity 0.90)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.89)
- EVODiff: Entropy-aware Variance Optimized Diffusion Inference (NeurIPS 2025 · similarity 0.89)
- Improved off-policy training of diffusion samplers (NeurIPS 2024 · similarity 0.89)
- Fast Solvers for Discrete Diffusion Models: Theory and Applications of High-Order Algorithms (NeurIPS 2025 · similarity 0.88)
- Towards Non-Asymptotic Convergence for Diffusion-Based Generative Models (ICLR 2024 · similarity 0.88)
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.
