Math Graduate Student Seminar

I will present two methodological advances at the interface of flow matching and optimal transport. First, I generalize Schrödinger bridges to multivariate Ornstein–Uhlenbeck reference processes, enabling efficient modelling of non-equilibrium and irreversible dynamics. Using closed-form OU bridges, this yields a simulation-free flow matching algorithm applicable to systems such as gene regulatory circuits and single-cell trajectories. Second, I show how semidiscrete optimal transport can accelerate generative modelling by providing low-variance, sample-efficient flow matching objectives. Together, these results demonstrate how structured reference dynamics and computational OT can improve the speed and fidelity of modern flow-based generative models.