Search papers, labs, and topics across Lattice.
This paper introduces a unified variational framework for Invertible Neural Networks (INNs) and Normalizing Flows (NFs) using unsupervised loss functions, drawing inspiration from GANs and Precision-Recall divergence. The work addresses the lack of theoretical guarantees on approximation quality in INNs and NFs under realistic assumptions. It derives theoretical performance guarantees for posterior accuracy in INNs and distributional accuracy in NFs under weaker assumptions than previous work, and validates the approach on an ocean-acoustic inversion problem.
VINA provides the first theoretical guarantees for INNs and NFs under realistic assumptions, finally bridging the gap between theory and practice for these powerful generative models.
The distinctive architectural features of normalizing flows (NFs), notably bijectivity and tractable Jacobians, make them well-suited for generative modeling. Invertible neural networks (INNs) build on these principles to address supervised inverse problems, enabling direct modeling of both forward and inverse mappings. In this paper, we revisit these architectures from both theoretical and practical perspectives and address a key gap in the literature: the lack of theoretical guarantees on approximation quality under realistic assumptions, whether for posterior inference in INNs or for generative modeling with NFs. We introduce a unified framework for INNs and NFs based on variational unsupervised loss functions, inspired by analogous formulations in related areas such as generative adversarial networks (GANs) and the Precision-Recall divergence for training normalizing flows. Within this framework, we derive theoretical performance guarantees, quantifying posterior accuracy for INNs and distributional accuracy for NFs, under assumptions that are weaker and more practically realistic than those used in prior work. Building on these theoretical results, we conduct extensive case studies to distill general design principles and practical guidelines. We conclude by demonstrating the effectiveness of our approach on a realistic ocean-acoustic inversion problem.
