StochasticInterpolantScheduler<T>

Models & Types · AiDotNet.Diffusion.Schedulers

Stochastic Interpolant scheduler for generalized flow-based sampling with noise injection.

Inheritance: NoiseSchedulerBase<T> → StochasticInterpolantScheduler<T>

For Beginners

This is a flexible sampler that can behave like either a flow model (deterministic, straight paths) or a diffusion model (stochastic, with randomness), or anything in between. You can tune the “stochasticity dial” to find the sweet spot.

How It Works

Implements the Stochastic Interpolant framework which unifies flow matching and score-based diffusion. Uses time-dependent interpolation between data and noise with optional stochasticity controlled by an auxiliary noise schedule.

Reference: Albergo et al., “Stochastic Interpolants: A Unifying Framework for Flows and Diffusions”, 2023

Constructors

Constructor Summary
StochasticInterpolantScheduler(SchedulerConfig<T> config, double stochasticity)

Methods

Method Summary
Vector<T> Step(Vector<T> modelOutput, int timestep, Vector<T> sample, T eta, Vector<T> noise)