English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Invertible neural networks (INNs) have been used to design generative models,implement memory-saving gradient computation, and solve inverse problems. Inthis work, we show that commonly-used INN architectures suffer from explodinginverses and are thus prone to becoming numerically non-invertible. Across awide range of INN use-cases, we reveal failures including the non-applicabilityof the change-of-variables formula on in- and out-of-distribution (OOD) data,incorrect gradients for memory-saving backprop, and the inability to samplefrom normalizing flow models. We further derive bi-Lipschitz properties ofatomic building blocks of common architectures. These insights into thestability of INNs then provide ways forward to remedy these failures. For taskswhere local invertibility is sufficient, like memory-saving backprop, wepropose a flexible and efficient regularizer. For problems where globalinvertibility is necessary, such as applying normalizing flows on OOD data, weshow the importance of designing stable INN building blocks.

Understanding and Mitigating Exploding Inverses in Invertible Neural Networks