BridgeTwist: Twisting Schrödinger Bridges for Training-Free Conditional Sampling
We propose BridgeTwist, a training-free conditional sampler for pretrained SB generative models, formulated as a tempered twisted sequential Monte Carlo scheme.
ProofPaper ↗
Key points
- Generative models are often required to serve many downstream conditional tasks.
- In the training-free setting, a pretrained unconditional model is reused as a prior and combined with a task-specific likelihood to sample from the corresponding posterior.
- While such training-free samplers have been developed for diffusion and flow models, their counterpart for Schrödinger bridge (SB) models remains unexplored, despite the growing appeal of SBs as flexible stochastic transport models between general source and target distributions.
- We further introduce a bridge-shaped inference-time diffusion schedule that concentrates the proposal noise in the interior of the path.
Sources (1)
- [1]BridgeTwist: Twisting Schrödinger Bridges for Training-Free Conditional SamplingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 04:20 PM
We propose BridgeTwist, a training-free conditional sampler for pretrained SB generative models, formulated as a tempered twisted sequential Monte Carlo scheme.
Generative models are often required to serve many downstream conditional tasks.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026Gemini 4 Argon for Robotics
- Oct 13, 2026LinMorph: Learning to Morph Weights for Efficient Video DiT Linearization
- Oct 13, 2026Risk-Averse Inference-Time Diffusion Model Alignment with Finite-Sample Guarantees
- Oct 13, 2026Scaling Online Optimal Transport Minibatches for Flow Matching
- Oct 13, 2026Efficient Scene-Text Refinement via Pixel-Space Grouped Image Generation
- Oct 12, 2026Image Tokenizers Should Be Trained Between Images