analiza aiqbench
This paper proposes a training-free way to steer pretrained Schrödinger bridge models toward conditional targets, such as class labels, text prompts, and inverse problems like inpainting and deblurring, without retraining. Its core technique is a tempered, twisted sequential Monte Carlo sampler with telescoping importance weights that the authors show is asymptotically exact. For readers, the main question is whether SB models can match the conditional-sampling toolkit already built for diffusion and flow models, and the reported results suggest they can, though the evidence so far is benchmark-level and the asymptotic guarantee depends on the approximation quality of the plug-in surrogate.
BridgeTwist: Twisting Schr\"odinger Bridges for Training-Free Conditional Sampling
arXiv:2610.18658v1 Announce Type: new Abstract: 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 Schr\"odinger bridge (SB) models remains unexplored, despite the growing appeal of SBs as flexible stochastic transport models between general source and target distributions. We propose BridgeTwist, a training-free conditional sampler for pretrained SB generative models, formulated as a tempered twisted sequential Monte Carlo scheme. Its twist is a closed-form plug-in surrogate, obtained by combining the velocity and score fields exposed by SB pretraining into an endpoint predictor and evaluating the likelihood there. Since this surrogate is less reliable near the source, we temper the twist with a time-varying exponent, downweighting its influence at early steps. We further introduce a bridge-shaped inference-time diffusion schedule that concentrates the proposal noise in the interior of the path. Telescoping importance weights then cancel all intermediate twist factors, yielding an asymptotically exact sampler for the posterior. Empirically, BridgeTwist outperforms training-free baselines on class-conditional and text-prompted image generation, and is competitive with specialized inverse problem solvers on inpainting, deblurring, and super-resolution.
Abstract: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 propose BridgeTwist, a training-free conditional sampler for pretrained SB generative models, formulated as a tempered twisted sequential Monte Carlo scheme. Its twist is a closed-form plug-in surrogate, obtained by combining the velocity and score fields exposed by SB pretraining into an endpoint predictor and evaluating the likelihood there. Since this surrogate is less reliable near the source, we temper the twist with a time-varying exponent, downweighting its influence at early steps. We further introduce a bridge-shaped inference-time diffusion schedule that concentrates the proposal noise in the interior of the path. Telescoping importance weights then cancel all intermediate twist factors, yielding an asymptotically exact sampler for the posterior. Empirically, BridgeTwist outperforms training-free baselines on class-conditional and text-prompted image generation, and is competitive with specialized inverse problem solvers on inpainting, deblurring, and super-resolution.
Machine Learning (cs.LG); Machine Learning (stat.ML)
arXiv:2610.18658 [cs.LG]
(or arXiv:2610.18658v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.18658 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 303 cuvinte
Subiecte conexe
SBNO : Schr\"odinger Bridge Neural Operator for the Forward and Inverse Problems under Missing Information
arXiv · 12 oct. 2026
Self-supervised In-context Operator Learning for Stochastic Mean-Field Control
arXiv · 20 aug. 2026
The Loss Floor of Denoising Score Matching: Fisher Geometry from Schr\"odinger Bridges
arXiv · 26 aug. 2026
Data-to-Energy Stochastic Dynamics
arXiv · 1 sept. 2026