NeurIPS 2020

Exchangeable Neural ODE for Set Modeling

Meta Review

Good paper that develops continuous-time normalizing flows for sets, by building in the associated permutation symmetry. Good contribution, that extends the field of normalizing flows into the direction of explicit incorporation of known symmetries. Well done. One concern the reviewers raised was that the Transformer-parameterized vector field may not be Lipschitz. I would encourage the authors to acknowledge this point in the camera-ready, and discuss whether we would expect this to be an issue or not.