1

We show that standard permutation equivariant denoisers cause severe limitations on such tasks, a problem that we pinpoint to their inability to break symmetries present in the noisy inputs.

we propose a generative model for antibody design using conjoined interacting neural ODEs

We propose generative graph normalizing flow models, based on a system of coupled node ODEs, that repeatedly reconcile locally toward globally aligned densities for high quality molecular generation.

We analyze the representational power and limits of modern models for (event-based) temporal graphs. We leverage our theoretical insights to introduce an architecture that is provably more expressive than existing ones.

A new formalism for disentanglement on graphs.