Zoloflow
A sampling algorithm that combines elliptic scheduling with customized discretization, based on Zolotarev optimal rational function approximation, which minimizes approximation error for a given number of function evaluations.
Timeline (last 90 days)
The research proposes a new framework based on the analytic properties of ODE trajectories, deriving a family of elliptic schedules via conformal mapping and, paired with a customized discretization, obtaining the Zoloflow algorithm.
The Zoloflow algorithm combines the elliptic schedule with a customized discretization that is closely related to the Zolotarev optimal rational function approximation and can minimize the approximation error for a given number of function evaluations.
For Gaussian targets, Zoloflow achieves minimax optimality among all oblivious score-span algorithms, requiring exactly Θ(logκ log(√d/ε)) exact score queries to sample with total variation error ε.
Oblivious score-span algorithms refer to sampling algorithms whose step-size sequence and query points do not depend on the real-time output of the score function; Zoloflow belongs to this class, so its complexity lower bound is universal.
The authors conduct experiments on synthetic benchmarks and existing image generation datasets, verifying that the elliptic schedule yields higher sampling quality when the condition number is large and is more computationally efficient when paired with Zoloflow.
All Facts (5)
The research proposes a new framework based on the analytic properties of ODE trajectories, deriving a family of elliptic schedules via conformal mapping and, paired with a customized discretization, obtaining the Zoloflow algorithm.
50%UnverifiedThe Zoloflow algorithm combines the elliptic schedule with a customized discretization that is closely related to the Zolotarev optimal rational function approximation and can minimize the approximation error for a given number of function evaluations.
50%UnverifiedFor Gaussian targets, Zoloflow achieves minimax optimality among all oblivious score-span algorithms, requiring exactly Θ(logκ log(√d/ε)) exact score queries to sample with total variation error ε.
50%UnverifiedOblivious score-span algorithms refer to sampling algorithms whose step-size sequence and query points do not depend on the real-time output of the score function; Zoloflow belongs to this class, so its complexity lower bound is universal.
50%UnverifiedThe authors conduct experiments on synthetic benchmarks and existing image generation datasets, verifying that the elliptic schedule yields higher sampling quality when the condition number is large and is more computationally efficient when paired with Zoloflow.
50%