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

The deep learning approach to detecting malicious software (malware) ispromising but has yet to tackle the problem of dataset shift, namely that thejoint distribution of examples and their labels associated with the test set isdifferent from that of the training set. This problem causes the degradation ofdeep learning models without users' notice. In order to alleviate the problem,one approach is to let a classifier not only predict the label on a givenexample but also present its uncertainty (or confidence) on the predictedlabel, whereby a defender can decide whether to use the predicted label or not.While intuitive and clearly important, the capabilities and limitations of thisapproach have not been well understood. In this paper, we conduct an empiricalstudy to evaluate the quality of predictive uncertainties of malware detectors.Specifically, we re-design and build 24 Android malware detectors (bytransforming four off-the-shelf detectors with six calibration methods) andquantify their uncertainties with nine metrics, including three metrics dealingwith data imbalance. Our main findings are: (i) predictive uncertainty indeedhelps achieve reliable malware detection in the presence of dataset shift, butcannot cope with adversarial evasion attacks; (ii) approximate Bayesian methodsare promising to calibrate and generalize malware detectors to deal withdataset shift, but cannot cope with adversarial evasion attacks; (iii)adversarial evasion attacks can render calibration methods useless, and it isan open problem to quantify the uncertainty associated with the predictedlabels of adversarial examples (i.e., it is not effective to use predictiveuncertainty to detect adversarial examples).

Can We Leverage Predictive Uncertainty to Detect Dataset Shift and Adversarial Examples in Android Malware Detection?