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

Data augmentation has emerged as a powerful technique for improving theperformance of deep neural networks and led to state-of-the-art results incomputer vision. However, state-of-the-art data augmentation strongly distortstraining images, leading to a disparity between examples seen during trainingand inference. In this work, we explore a recently proposed training paradigmin order to correct for this disparity: using an auxiliary BatchNorm for thepotentially out-of-distribution, strongly augmented images. Our experimentsthen focus on how to define the BatchNorm parameters that are used atevaluation. To eliminate the train-test disparity, we experiment with using thebatch statistics defined by clean training images only, yet surprisingly findthat this does not yield improvements in model performance. Instead, weinvestigate using BatchNorm parameters defined by weak augmentations and findthat this method significantly improves the performance of common imageclassification benchmarks such as CIFAR-10, CIFAR-100, and ImageNet. We thenexplore a fundamental trade-off between accuracy and robustness coming fromusing different BatchNorm parameters, providing greater insight into thebenefits of data augmentation on model performance.

Does Data Augmentation Benefit from Split BatchNorms