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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Background: When using deep learning models, there are many possiblevulnerabilities and some of the most worrying are the adversarial inputs, whichcan cause wrong decisions with minor perturbations. Therefore, it becomesnecessary to retrain these models against adversarial inputs, as part of thesoftware testing process addressing the vulnerability to these inputs.Furthermore, for an energy efficient testing and retraining, data scientistsneed support on which are the best guidance metrics and optimal datasetconfigurations.  Aims: We examined four guidance metrics for retraining convolutional neuralnetworks and three retraining configurations. Our goal is to improve the modelsagainst adversarial inputs regarding accuracy, resource utilization and timefrom the point of view of a data scientist in the context of imageclassification.  Method: We conducted an empirical study in two datasets for imageclassification. We explore: (a) the accuracy, resource utilization and time ofretraining convolutional neural networks by ordering new training set by fourdifferent guidance metrics (neuron coverage, likelihood-based surpriseadequacy, distance-based surprise adequacy and random), (b) the accuracy andresource utilization of retraining convolutional neural networks with threedifferent configurations (from scratch and augmented dataset, using weights andaugmented dataset, and using weights and only adversarial inputs).  Results: We reveal that retraining with adversarial inputs from originalweights and by ordering with surprise adequacy metrics gives the best modelw.r.t. the used metrics.  Conclusions: Although more studies are necessary, we recommend datascientists to use the above configuration and metrics to deal with thevulnerability to adversarial inputs of deep learning models, as they canimprove their models against adversarial inputs without using many inputs.

Guiding the retraining of convolutional neural networks against adversarial inputs