This page is under construction
The Schrödinger Bridge Problem
The Schrödinger Bridge Problem is a weak version of the Optimal Transport
Flow Matching and The Conditioning Trick
Flow Matching[1] introduce the conditioning trick. The conditioning trick is first derived in the context of speeding up continuous flow model but its ability for allowing simulation-free training is later used in training Schrödinger Bridges[3]. In short, the conditioning trick can be seen in the simplification of the flow-matching loss:
The Diffusion Process
The diffusion equation or also known as the Fokker-Planck equation is described by a hyperbolic PDE:
This diffusion process can also be described by the corresponding SDE:
The link between these two equation can be seen through the proof of the famous Feynmann-Kac’s formula.
References
[0] SB-FBSDE: Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory
[1] Flow Matching for Generative Modeling
[2] I2SB: Image-to-Image Schrödinger Bridge