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 Neural Network (DNN) trained by the gradient descent method is known tobe vulnerable to maliciously perturbed adversarial input, aka. adversarialattack. As one of the countermeasures against adversarial attack, increasingthe model capacity for DNN robustness enhancement was discussed and reported asan effective approach by many recent works. In this work, we show thatshrinking the model size through proper weight pruning can even be helpful toimprove the DNN robustness under adversarial attack. For obtaining asimultaneously robust and compact DNN model, we propose a multi-objectivetraining method called Robust Sparse Regularization (RSR), through the fusionof various regularization techniques, including channel-wise noise injection,lasso weight penalty, and adversarial training. We conduct extensiveexperiments across popular ResNet-20, ResNet-18 and VGG-16 DNN architectures todemonstrate the effectiveness of RSR against popular white-box (i.e., PGD andFGSM) and black-box attacks. Thanks to RSR, 85% weight connections of ResNet-18can be pruned while still achieving 0.68% and 8.72% improvement in clean- andperturbed-data accuracy respectively on CIFAR-10 dataset, in comparison to itsPGD adversarial training baseline.

Robust Sparse Regularization: Simultaneously Optimizing Neural Network Robustness and Compactness