Research map / Image generation / Fast diffusion sampling
Stable Target Field for Reduced Variance Score Estimation in Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DSM with STF for EDM using NCSN++ | EDM, NCSN++ · CIFAR-10 | FID: 1.98 -> 1.90 | Helped |
| Replace DSM with STF for EDM using DDPM++ | EDM, DDPM++ · CIFAR-10 | FID: 1.97 -> 1.92 | Helped |
| Replace DSM with STF for VE using RK45 | VE · CIFAR-10 | FID: 8.90 -> 5.51 | Helped |
| Replace DSM with STF for VE using PC | VE · CIFAR-10 | FID: 2.75 -> 2.66 | Helped |
| Replace DSM with STF for VP using DDIM | VP · CIFAR-10 | FID: 5.16 -> 5.06 | Helped |
| Replace DSM with STF for VP using RK45 | VP · CIFAR-10 | FID: 2.90 -> 2.99 | Hurt |
| Replace DSM with STF for EDM using NCSN++ | EDM, NCSN++ · CIFAR-10 | Inception: 9.82 -> 9.93 | Helped |
| Replace DSM with STF for EDM using DDPM++ | EDM, DDPM++ · CIFAR-10 | Inception: 9.78 -> 9.79 | Helped |
| Replace DSM with STF for VE using RK45 | VE · CIFAR-10 | Inception: 9.27 -> 9.52 | Helped |
| Replace DSM with STF for VP using RK45 | VP · CIFAR-10 | Inception: 9.46 -> 9.43 | Hurt |
| Use a large reference batch for STF instead of a single training target | STF · CIFAR-10 | training time to comparable FID: 3.6× speed-up; 25h versus 90h | Helped |
| Increase the STF reference batch size | STF · CIFAR-10 | FID | Helped |
Closest papers by meaning
- Generalization through variance: how noise shapes inductive biases in diffusion models (ICLR 2025 · similarity 0.91)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.91)
- Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers (ICLR 2025 · similarity 0.91)
- Adapting to Unknown Low-Dimensional Structures in Score-Based Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Consistent Diffusion Meets Tweedie: Training Exact Ambient Diffusion Models with Noisy Data (ICML 2024 · similarity 0.90)
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.90)
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.