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

Data Poisoning attacks modify training data to maliciously control a modeltrained on such data. In this work, we focus on targeted poisoning attackswhich cause a reclassification of an unmodified test image and as such breachmodel integrity. We consider a particularly malicious poisoning attack that isboth "from scratch" and "clean label", meaning we analyze an attack thatsuccessfully works against new, randomly initialized models, and is nearlyimperceptible to humans, all while perturbing only a small fraction of thetraining data. Previous poisoning attacks against deep neural networks in thissetting have been limited in scope and success, working only in simplifiedsettings or being prohibitively expensive for large datasets. The centralmechanism of the new attack is matching the gradient direction of maliciousexamples. We analyze why this works, supplement with practical considerations.and show its threat to real-world practitioners, finding that it is the firstpoisoning method to cause targeted misclassification in modern deep networkstrained from scratch on a full-sized, poisoned ImageNet dataset. Finally wedemonstrate the limitations of existing defensive strategies against such anattack, concluding that data poisoning is a credible threat, even forlarge-scale deep learning systems.

Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching