Universal approximation property of invertible neural networks

Invertible neural networks (INNs) are neural network architectures with invertibility by design. Thanks to their invertibility and the tractability of Jacobian, INNs have various machine learning applications such as probabilistic modeling, generative modeling, and representation learning. However, their attractive properties often come at the cost of restricting the layer designs, which poses a question on their representation power: can we use these models to approximate sufficiently diverse functions? To answer this question, we have developed a general theoretical framework to investigate the representation power of INNs, building on a structure theorem of differential geometry. The framework simplifies the approximation problem of diffeomorphisms, which enables us to show the universal approximation properties of INNs. We apply the framework to two representative classes of INNs, namely Coupling-Flow-based INNs (CF-INNs) and Neural Ordinary Differential Equations (NODEs), and elucidate their high representation power despite the restrictions on their architectures.
Comment: This paper extends our previous work of the following two papers: "Coupling-based invertible neural networks are universal diffeomorphism approximators" [arXiv:2006.11469] (published as a conference paper in NeurIPS 2020) and "Universal approximation property of neural ordinary differential equations" [arXiv:2012.02414] (presented at DiffGeo4DL Workshop in NeurIPS 2020)