On the Universality of Augmented Invertible Networks
Keywords: Invertible networks, pseudo-invertible, hamiltonian, symplectic, coupling layer, universality
Abstract: We revisit the universality of augmented invertible networks, reversible architectures that leverage zero padding to increase their expressiveness. Under mild hypotheses, we provide a short proof that a single augmented RevNet block is an universal approximator for Bi-Lipschitz homeomorphisms, which we then extend to Augmented Neural ODEs and i-ResNets. We demonstrate that augmenting these architectures with additional dimensions essentially trivializes their invertible structure, rendering them equivalent to encoder-decoder ensembles. We embrace this equivalence and argue for the use of augmented RevNets, which we demonstrate to be both faster and more expressive in low-dimensional numerical experiments, as compared to other approximately invertible models.
Primary Area: learning theory
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8111
Loading