Research map / Probabilistic and scientific modeling / Variational and generative sampling
Discrete Adjoint Schrödinger Bridge Sampler
Wei Guo, Yuchen Zhu, Xiaochen Du, Juno Nam, Yongxin Chen, Rafael Gomez-Bombarelli, Guan-Horng Liu, Molei Tao, 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 discrete ASBS instead of LEAPS | discrete neural sampler · Ising L=24 beta_high | training runtime (h): 8.4 -> 0.2 | Improved |
| use discrete ASBS instead of MDNS | discrete neural sampler · Ising L=24 beta_high | training runtime (h): 16.8 -> 0.2 | Improved |
| use discrete ASBS instead of LEAPS at critical temperature | uniform-based sampler · Ising L=24 beta_critical | EW2: 96.5 -> 53.7 | Improved |
| use discrete ASBS instead of LEAPS at low temperature | uniform-based sampler · Ising L=24 beta_low | EW2: 176.6 -> 2.5 | Improved |
| use discrete ASBS instead of LEAPS at low temperature | uniform-based sampler · Potts L=16 beta_low | EW2: 90.5 -> 1.4 | Improved |
| use discrete ASBS instead of masked MDNS at high temperature | sampler · Ising L=24 beta_high | EW2: 0.1 -> 5.4 | Hurt |
| use discrete ASBS instead of MDNS at critical temperature | sampler · Potts L=16 beta_critical | EW2: 1.9 -> 14.6 | Hurt |
| use adjoint matching loss instead of denoising matching for the controller (ablation) | DASBS controller · Potts L=8 N=3 beta_high=0.5 | EW2 convergence speed | Improved |
| add trajectory importance reweighting to training losses (ablation) | DASBS · Potts L=8 N=3 beta_high=0.5 | convergence speed | Hurt |
| use non-memoryless noise schedule (alpha > 0) instead of memoryless (alpha = 0) (ablation) | DASBS · Ising L=24 beta_high=0.28 | 2-pt correlation error and EW2 | Improved |
| vary gamma or alpha of schedule gamma/(t+alpha) away from optimum (ablation) | DASBS · Ising L=24 beta_high=0.28 | 2-pt correlation error and EW2 | Mixed |
| use Metropolis-Hastings baseline at critical temperature | MH · Ising L=24 beta_critical | EW2: 53.7 -> 293.3 | Hurt |
Closest papers by meaning
- Adjoint Schrödinger Bridge Sampler (NeurIPS 2025 · similarity 0.91)
- Generalized Schrödinger Bridge Matching (ICLR 2024 · similarity 0.87)
- Momentum Multi-Marginal Schrödinger Bridge Matching (NeurIPS 2025 · similarity 0.87)
- Categorical Schrödinger Bridge Matching (ICML 2025 · similarity 0.87)
- Adjoint Sampling: Highly Scalable Diffusion Samplers via Adjoint Matching (ICML 2025 · similarity 0.87)
- Discrete Diffusion Schrödinger Bridge Matching for Graph Transformation (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.
