We propose Permutation Equivariant Graph Neural CDEs, an equivariant and parameter-efficient extension of Graph Neural CDEs for dynamic graph representation learning.

We present a theoretical framework which unifies posterior and end-to-end guidance for flow/diffusion models.