Research map / Image generation / Fast diffusion sampling
Trust-Region Noise Search for Black-Box Alignment of Diffusion and Flow Models
Niklas Schweiger, Karnik Ram, Daniel Cremers
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 trust-region side-length adaptation (ablation) | TRS · DrawBench (SD1.5) | mean best ImageReward: 1.60 -> 1.62 | Improved |
| add trust-region side-length adaptation (ablation) | TRS · DrawBench (SD1.5) | mean best HPSv2: 0.318 -> 0.322 | Improved |
| add trust-region side-length adaptation (ablation) | TRS · DrawBench (SDXL-Lightning) | mean best ImageReward: 1.64 -> 1.66 | Improved |
| add trust-region side-length adaptation (ablation) | TRS · DrawBench (SDXL-Lightning) | mean best HPSv2: 0.333 -> 0.340 | Improved |
| replace random search with TRS | source-noise search · DrawBench (SD1.5) | mean best ImageReward: 1.44 -> 1.62 | Improved |
| replace random search with TRS | source-noise search · DrawBench (SDXL-Lightning) | mean best HPSv2: 0.324 -> 0.340 | Improved |
| replace DTS* with TRS (fewer reward calls) | inference-time alignment · DrawBench (SD1.5) | wall-clock time per prompt (s): 924.9 -> 226.5 | Improved |
| replace DTS* with TRS | inference-time alignment · DrawBench (SD1.5) | mean best ImageReward: 1.59 -> 1.62 | Improved |
| replace diagonal CMA-ES with TRS | source-noise search · DrawBench (SDXL-Lightning) | mean best ImageReward: 1.67 -> 1.66 | No clear effect |
| replace diagonal CMA-ES with TRS | source-noise search · DrawBench (SD1.5) | mean best HPSv2: 0.319 -> 0.322 | Improved |
| replace OC-Flow gradient-based optimization with Fast Direct or random search | noise optimization · DrawBench (SD1.5) | mean best ImageReward: 0.42 -> 1.35 | Improved |
| replace OC-Flow with TRS (memory and time) | gradient-based guidance · DrawBench (SD1.5) | wall-clock time per prompt (s): 335 -> 226.5 | Improved |
Closest papers by meaning
- Inference-Time Alignment of Diffusion Models with Direct Noise Optimization (ICML 2025 · similarity 0.89)
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-based Decoding (NeurIPS 2025 · similarity 0.88)
- Steering Masked Discrete Diffusion Models via Discrete Denoising Posterior Prediction (ICLR 2025 · similarity 0.88)
- Reward-Guided Iterative Refinement in Diffusion Models at Test-Time with Applications to Protein and DNA Design (ICML 2025 · similarity 0.87)
- Diffusion Tree Sampling: Scalable inference‑time alignment of diffusion models (NeurIPS 2025 · similarity 0.87)
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.
