aiqbench analysis
This arXiv paper proposes an "any-step" stochastic flow map for generative models built on stochastic differential equations, extending one-step distillation methods that have so far relied on deterministic ODEs. Its practical claim is that the map gives cheap, differentiable access to posterior samples, which could make inference-time steering and conditional sampling easier to do. The reported results are on synthetic and image-generation benchmarks, so readers should treat the gains as early evidence rather than a settled improvement over existing samplers.
It\^o maps for any-step SDEs
arXiv:2606.11156v2 Announce Type: replace-cross Abstract: Recent one-step generative models accelerate sampling by learning deterministic flow maps of the underlying dynamics. These methods rely on learning from ordinary differential equations, leaving open how to define an exact distillation procedure for stochastic dynamics. We introduce the It\^o map, an any-step stochastic flow map that takes an intermediate state and Brownian path and predicts future states in a single pass. The It\^o map formulation yields novel estimators for inference-time control by providing cheap, differentiable access to posterior samples. Empirically, It\^o maps produce diverse, conditionally valid endpoint samples from fixed intermediate states and support strong steering performance on synthetic and image-generation benchmarks. These results establish any-step SDE integration as a useful primitive for posterior sampling and stochastic control.
Abstract:Recent one-step generative models accelerate sampling by learning deterministic flow maps of the underlying dynamics. These methods rely on learning from ordinary differential equations, leaving open how to define an exact distillation procedure for stochastic dynamics. We introduce the Itô map, an any-step stochastic flow map that takes an intermediate state and Brownian path and predicts future states in a single pass. The Itô map formulation yields novel estimators for inference-time control by providing cheap, differentiable access to posterior samples. Empirically, Itô maps produce diverse, conditionally valid endpoint samples from fixed intermediate states and support strong steering performance on synthetic and image-generation benchmarks. These results establish any-step SDE integration as a useful primitive for posterior sampling and stochastic control.
Machine Learning (stat.ML); Machine Learning (cs.LG)
arXiv:2606.11156 [stat.ML]
(or arXiv:2606.11156v2 [stat.ML] for this version)
https://doi.org/10.48550/arXiv.2606.11156 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 214 words