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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep neural networks often exhibit poor performance on data that is unlikelyunder the train-time data distribution, for instance data affected bycorruptions. Previous works demonstrate that test-time adaptation to datashift, for instance using entropy minimization, effectively improvesperformance on such shifted distributions. This paper focuses on the fullytest-time adaptation setting, where only unlabeled data from the targetdistribution is required. This allows adapting arbitrary pretrained networks.Specifically, we propose a novel loss that improves test-time adaptation byaddressing both premature convergence and instability of entropy minimization.This is achieved by replacing the entropy by a non-saturating surrogate andadding a diversity regularizer based on batch-wise entropy maximization thatprevents convergence to trivial collapsed solutions. Moreover, we propose toprepend an input transformation module to the network that can partially undotest-time distribution shifts. Surprisingly, this preprocessing can be learnedsolely using the fully test-time adaptation loss in an end-to-end fashionwithout any target domain labels or source domain data. We show that ourapproach outperforms previous work in improving the robustness of publiclyavailable pretrained image classifiers to common corruptions on suchchallenging benchmarks as ImageNet-C.

Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformation