Research map / Probabilistic and scientific modeling / Bayesian optimization
Generative Modeling from Black-Box Corruptions via Self-Consistent Stochastic Interpolants
Chirag Modi, Jiequn Han, Eric Vanden-Eijnden, Joan Bruna
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 SCSI instead of DPS (trained on clean data) | SCSI · Gaussian blur sigma_n=0.25 | LPIPS: 0.025 -> 0.015 | Improved |
| use SCSI instead of DPS at low noise | SCSI · Motion blur sigma_n=1e-6 | LPIPS: 0.0026 -> 0.0069 | Hurt |
| use SCSI instead of Ambient Diffusion | SCSI · CIFAR-10 random masking | FID: 18.85 -> 6.74 (rho 0.4 vs 0.5) | Improved |
| use SCSI instead of EM Posterior | SCSI · CIFAR-10 random masking | FID: 6.76 -> 6.74 (rho=0.50) | No clear effect |
| use SDE instead of ODE transport at high noise (ablation) | SCSI · Two-moon, sigma_n=1.0 | restoration stability | Improved |
| use SDE instead of ODE in high-dimensional tasks (ablation) | SCSI · imaging | hyperparameter sensitivity | Hurt |
| increase inner-loop transport steps T_tr from 1 to 1000 (ablation) | SCSI · Two-moon AWGN | Wasserstein distance: 0.0491 -> 0.0593 | Hurt |
| increase T_tr from 1 to 10 (ablation) | SCSI · Two-moon AWGN | Wasserstein distance: 0.0491 -> 0.0460 | No clear effect |
Closest papers by meaning
- Inverse Flow and Consistency Models (ICML 2025 · similarity 0.89)
- Stochastic Interpolants with Data-Dependent Couplings (ICML 2024 · similarity 0.88)
- Stochastic Forward–Backward Deconvolution: Training Diffusion Models with Finite Noisy Datasets (ICML 2025 · similarity 0.88)
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.88)
- Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations (ICLR 2025 · similarity 0.87)
- DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform (NeurIPS 2024 · 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.
