Generative Modeling via Nonlinear Schr\"odinger Bridges and Branching Diffusions

The SBNL framework uses coupled forward-backward partial differential equations from mean-field game theory to solve nonlinear Schr\"odinger bridge problems. Implementation relies on a branching diffusion simulation algorithm that leverages the probabilistic representation of semilinear PDEs. This approach provides an embarrassingly parallel method for modeling complex nonlinear interaction functionals in high dimensions.