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 learning (DL) plays a more and more important role in our daily life dueto its competitive performance in industrial application domains. As the coreof DL-enabled systems, deep neural networks (DNNs) need to be carefullyevaluated to ensure the produced models match the expected requirements. Inpractice, the \emph{de facto standard} to assess the quality of DNNs in theindustry is to check their performance (accuracy) on a collected set of labeledtest data. However, preparing such labeled data is often not easy partlybecause of the huge labeling effort, i.e., data labeling is labor-intensive,especially with the massive new incoming unlabeled data every day. Recentstudies show that test selection for DNN is a promising direction that tacklesthis issue by selecting minimal representative data to label and using thesedata to assess the model. However, it still requires human effort and cannot beautomatic. In this paper, we propose a novel technique, named \textit{Aries},that can estimate the performance of DNNs on new unlabeled data using only theinformation obtained from the original test data. The key insight behind ourtechnique is that the model should have similar prediction accuracy on the datawhich have similar distances to the decision boundary. We performed alarge-scale evaluation of our technique on two famous datasets, CIFAR-10 andTiny-ImageNet, four widely studied DNN models including ResNet101 andDenseNet121, and 13 types of data transformation methods. Results show that theestimated accuracy by \textit{Aries} is only 0.03\% -- 2.60\% off the trueaccuracy. Besides, \textit{Aries} also outperforms the state-of-the-artlabeling-free methods in 50 out of 52 cases and selection-labeling-basedmethods in 96 out of 128 cases.

Aries: Efficient Testing of Deep Neural Networks via Labeling-Free Accuracy Estimation