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/

Zendy Plus

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

Zendy Tools

Zendy Open

Adversarial Training (AT), pivotal in fortifying the robustness of deeplearning models, is extensively adopted in practical applications. However,prevailing AT methods, relying on direct iterative updates for target model'sdefense, frequently encounter obstacles such as unstable training andcatastrophic overfitting. In this context, our work illuminates the potentialof leveraging the target model's historical states as a proxy to provideeffective initialization and defense prior, which results in a general proxyguided defense framework, `LAST' ({\bf L}earn from the P{\bf ast}).Specifically, LAST derives response of the proxy model as dynamically learnedfast weights, which continuously corrects the update direction of the targetmodel. Besides, we introduce a self-distillation regularized defense objective,ingeniously designed to steer the proxy model's update trajectory withoutresorting to external teacher models, thereby ameliorating the impact ofcatastrophic overfitting on performance. Extensive experiments and ablationstudies showcase the framework's efficacy in markedly improving modelrobustness (e.g., up to 9.2\% and 20.3\% enhancement in robust accuracy onCIFAR10 and CIFAR100 datasets, respectively) and training stability. Theseimprovements are consistently observed across various model architectures,larger datasets, perturbation sizes, and attack modalities, affirming LAST'sability to consistently refine both single-step and multi-step AT strategies.The code will be available at~\url{https://github.com/callous-youth/LAST}.

Learn from the Past: A Proxy Guided Adversarial Defense Framework with Self Distillation Regularization