Research map / Image generation / Fast diffusion sampling
Efficient Generative Modeling beyond Memoryless Diffusion via Adjoint Schrödinger Bridge Matching
Jeongwoo Shin, Jinhwan Sul, Joonseok Lee, Jaewoong Choi, Jaemoo Choi
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 ASBM instead of Score SDE | diffusion model · CIFAR-10 (100 NFE) | FID: 4.61 -> 3.16 | Improved |
| use ASBM instead of Score SDE at low NFE | diffusion model · FFHQ latent, NFE 25 | FID: 52.08 -> 8.85 | Improved |
| use ASBM instead of Score SDE at 1000 NFE | diffusion model · FFHQ latent, NFE 1000 | FID: 6.63 -> 6.27 | Improved |
| use non-memoryless base SDE in SB-FBSDE (ablation) | SB-FBSDE · CIFAR-10 across NFE | FID | Hurt |
| drop empirical bridge-matching pretraining (ablation) | DSBM · CIFAR-10 across NFE | FID | Hurt |
| use ASBM two-stage training instead of alternating forward-backward | SB-based generators · CIFAR-10 Heun 25 steps | FID: 39.84 -> 3.74 | Improved |
| use ASBM instead of Score SDE for trajectories | diffusion model · CIFAR-10 | trajectory straightness | Improved |
| use ASBM instead of Score SDE for drift | diffusion model · CIFAR-10 | drift directionality: 0.0602 -> 0.3672 | Improved |
| distill ASBM instead of score distillation (DMD) | one-step generator · CIFAR-10 | FID: 8.25 -> 6.68 | Improved |
| distill ASBM instead of DMD | one-step generator · CIFAR-10 | recall: 0.513 -> 0.542 | Improved |
| distill ASBM instead of DMD | one-step generator · CIFAR-10 | precision: 0.715 -> 0.702 | No clear effect |
| make base SDE too close to deterministic (small beta_max, e.g. 1) (ablation) | ASBM · CIFAR-10 | FID | Hurt |
Closest papers by meaning
- Generalized Schrödinger Bridge Matching (ICLR 2024 · similarity 0.91)
- Adjoint Schrödinger Bridge Sampler (NeurIPS 2025 · similarity 0.90)
- Schrodinger Bridge Flow for Unpaired Data Translation (NeurIPS 2024 · similarity 0.90)
- BridgeTwist: Twisting Schrödinger Bridges for Training-Free Conditional Sampling (NeurIPS 2026 · similarity 0.90)
- Adversarial Schrödinger Bridge Matching (NeurIPS 2024 · similarity 0.89)
- Momentum Multi-Marginal Schrödinger Bridge Matching (NeurIPS 2025 · similarity 0.89)
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.
