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

Backdoor data poisoning attacks have recently been demonstrated in computervision research as a potential safety risk for machine learning (ML) systems.Traditional data poisoning attacks manipulate training data to induceunreliability of an ML model, whereas backdoor data poisoning attacks maintainsystem performance unless the ML model is presented with an input containing anembedded "trigger" that provides a predetermined response advantageous to theadversary. Our work builds upon prior backdoor data-poisoning research for MLimage classifiers and systematically assesses different experimental conditionsincluding types of trigger patterns, persistence of trigger patterns duringretraining, poisoning strategies, architectures (ResNet-50, NasNet,NasNet-Mobile), datasets (Flowers, CIFAR-10), and potential defensiveregularization techniques (Contrastive Loss, Logit Squeezing, Manifold Mixup,Soft-Nearest-Neighbors Loss). Experiments yield four key findings. First, thesuccess rate of backdoor poisoning attacks varies widely, depending on severalfactors, including model architecture, trigger pattern and regularizationtechnique. Second, we find that poisoned models are hard to detect throughperformance inspection alone. Third, regularization typically reduces backdoorsuccess rate, although it can have no effect or even slightly increase it,depending on the form of regularization. Finally, backdoors inserted throughdata poisoning can be rendered ineffective after just a few epochs ofadditional training on a small set of clean data without affecting the model'sperformance.

Systematic Evaluation of Backdoor Data Poisoning Attacks on Image Classifiers