Stochastic Interpolant
A mathematical framework for constructing generative models that connects a noise distribution to a target data distribution via stochastic differential equation paths. The framework allows free choice of interpolation path endpoints and functional form. Flow matching is a special case corresponding to its deterministic variant.
Timeline (last 90 days)
Stochastic interpolators are a class of mathematical frameworks for constructing generative models that connect a noise distribution to a target data distribution via stochastic differential equation paths, allowing free choice of the endpoints and form of the interpolation path.
Flow Matching can be viewed as a special case of the deterministic version of stochastic interpolators.
All Facts (2)
Stochastic interpolators are a class of mathematical frameworks for constructing generative models that connect a noise distribution to a target data distribution via stochastic differential equation paths, allowing free choice of the endpoints and form of the interpolation path.
50%UnverifiedFlow Matching can be viewed as a special case of the deterministic version of stochastic interpolators.
50%