ott.neural.methods.flows.dynamics.BrownianBridge#
- class ott.neural.methods.flows.dynamics.BrownianBridge(sigma)[source]#
Brownian Bridge.
Sampler for sampling noise implicitly defined by a Schrödinger Bridge problem with parameter \(\sigma\) such that \(\sigma_t = \sigma \cdot \sqrt{t \cdot (1 - t)}\) [Tong et al., 2023].
- Parameters:
sigma (
float
) – Noise used for computing time-dependent noise schedule.
Methods
compute_mu_t
(t, src, tgt)Compute the mean of the probability path.
Compute noise of the flow at time \(t\).
compute_ut
(t, src, tgt)Evaluate the conditional vector field.
compute_xt
(rng, t, src, tgt)Sample from the probability path.