We leverage the closed-form formulation of flow matching to understand its generalization

We introduce Conformal Online Learning (COL), a novel framework for adaptively updating Koopman-invariant representations of nonlinear dynamical systems from streaming data.