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

Despite their unprecedented performance in various domains, utilization ofDeep Neural Networks (DNNs) in safety-critical environments is severely limitedin the presence of even small adversarial perturbations. The present workdevelops a randomized approach to detecting such perturbations based on minimumuncertainty metrics that rely on sampling at the hidden layers during the DNNinference stage. Inspired by Bayesian approaches to uncertainty estimation, thesampling probabilities are designed for effective detection of theadversarially corrupted inputs. Being modular, the novel detector ofadversaries can be conveniently employed by any pre-trained DNN at no extratraining overhead. Selecting which units to sample per hidden layer entailsquantifying the amount of DNN output uncertainty, where the overall uncertaintyis expressed in terms of its layer-wise components - what also promotesscalability. Sampling probabilities are then sought by minimizing uncertaintymeasures layer-by-layer, leading to a novel convex optimization problem thatadmits an exact solver with superlinear convergence rate. By simplifying theobjective function, low-complexity approximate solvers are also developed. Inaddition to valuable insights, these approximations link the novel approachwith state-of-the-art randomized adversarial detectors. The effectiveness ofthe novel detectors in the context of competing alternatives is highlightedthrough extensive tests for various types of adversarial attacks with variablelevels of strength.

Minimum Uncertainty Based Detection of Adversaries in Deep Neural Networks