Research map / Image generation / Fast diffusion sampling
Control Consistency Losses for Diffusion Bridges
Samuel Howard, Nikolas Nüsken, Jakiw Pidstrigach
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 self-consistency property SC2 instead of SC1 (ablation) | CCDB · Double Well (v = 3.0) | KL(P*||Ptheta): 0.148±0.073 -> 0.041±0.013 | Improved |
| use SC1 instead of SC2 (ablation) | CCDB · CIR model | KL(P*||Ptheta): 0.055±0.009 -> 0.023±0.010 | Improved |
| use SC1 vs SC2 in stable settings (ablation) | CCDB · Cell Diffusion (all three settings) | KL(Ptheta||P): 6.53 vs 6.53; 66.1 vs 66.0; 792.8 vs 792.4 | No clear effect |
| use SC2 instead of SC1 (ablation) | CCDB · Muller-Brown | learning the correct dynamics | Improved |
| train on controlled-dynamics samples (CCDB) instead of uncontrolled-dynamics methods | CCDB vs SDB/FB/BEL · Cell Diffusion (multimodal) | KL(Ptheta||P): 792.4±1.0 vs 885.1±6.0 (BEL) | Improved |
| replace NGDB (backprop through trajectories) with CCDB (no backprop) | CCDB · Training runtime, Muller-Brown (100d) | runtime: 5300s -> 1080s | Improved |
| replace NGDB with CCDB on accuracy | CCDB · Cell Diffusion (multimodal) | KL(Ptheta||P): 783.4±0.8 -> 792.4±1.0 | Hurt |
| use CCDB instead of NGDB | CCDB · Ornstein-Uhlenbeck bridges | KL to ground truth | Improved |
| use CCDB instead of Doob's Lagrangian | CCDB · Muller-Brown (2d) | KL(Ptheta||P): 38.6±0.2 -> 27.8±1.7 | Improved |
| use endpoint (final-step) alpha schedule instead of average schedule (ablation) | CCDB · Appendix E.7 experiments | performance | Hurt |
| add sticking-the-landing adjustments (ablation) | CCDB · Double Well | performance | Improved |
Closest papers by meaning
- Neural Guided Diffusion Bridges (ICML 2025 · similarity 0.90)
- Stochastic Optimal Control for Diffusion Bridges in Function Spaces (NeurIPS 2024 · similarity 0.90)
- Stochastic Control for Fine-tuning Diffusion Models: Optimality, Regularity, and Convergence (ICML 2025 · similarity 0.87)
- Conditioning Diffusions Using Malliavin Calculus (ICML 2025 · similarity 0.87)
- Elucidating the Preconditioning in Consistency Distillation (ICLR 2025 · similarity 0.87)
- Underdamped Diffusion Bridges with Applications to Sampling (ICLR 2025 · similarity 0.86)
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.
