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

Modern deep generative models can assign high likelihood to inputs drawn fromoutside the training distribution, posing threats to models in open-worlddeployments. While much research attention has been placed on defining newtest-time measures of OOD uncertainty, these methods do not fundamentallychange how deep generative models are regularized and optimized in training. Inparticular, generative models are shown to overly rely on the backgroundinformation to estimate the likelihood. To address the issue, we propose anovel frequency-regularized learning FRL framework for OOD detection, whichincorporates high-frequency information into training and guides the model tofocus on semantically relevant features. FRL effectively improves performanceon a wide range of generative architectures, including variationalauto-encoder, GLOW, and PixelCNN++. On a new large-scale evaluation task, FRLachieves the state-of-the-art performance, outperforming a strong baselineLikelihood Regret by 10.7% (AUROC) while achieving 147$\times$ faster inferencespeed. Extensive ablations show that FRL improves the OOD detection performancewhile preserving the image generation quality. Code is available athttps://github.com/mu-cai/FRL.

Out-of-distribution Detection via Frequency-regularized Generative Models