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

Making deep neural networks robust to small adversarial noises has recentlybeen sought in many applications. Adversarial training through iterativeprojected gradient descent (PGD) has been established as one of the mainstreamideas to achieve this goal. However, PGD is computationally demanding and oftenprohibitive in case of large datasets and models. For this reason, single-stepPGD, also known as FGSM, has recently gained interest in the field.Unfortunately, FGSM-training leads to a phenomenon called ``catastrophicoverfitting," which is a sudden drop in the adversarial accuracy under the PGDattack. In this paper, we support the idea that small input gradients play akey role in this phenomenon, and hence propose to zero the input gradientelements that are small for crafting FGSM attacks. Our proposed idea, whilebeing simple and efficient, achieves competitive adversarial accuracy onvarious datasets.

ZeroGrad : Mitigating and Explaining Catastrophic Overfitting in FGSM Adversarial Training