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

Malicious perturbations embedded in input data, known as Trojan attacks, cancause neural networks to misbehave. However, the impact of a Trojan attack isreduced during fine-tuning of the model, which involves transferring knowledgefrom a pretrained large-scale model like visual question answering (VQA) to thetarget model. To mitigate the effects of a Trojan attack, replacing andfine-tuning multiple layers of the pretrained model is possible. This researchfocuses on sample efficiency, stealthiness and variation, and robustness tomodel fine-tuning. To address these challenges, we propose an instance-levelTrojan attack that generates diverse Trojans across input samples andmodalities. Adversarial learning establishes a correlation between a specifiedperturbation layer and the misbehavior of the fine-tuned model. We conductedextensive experiments on the VQA-v2 dataset using a range of metrics. Theresults show that our proposed method can effectively adapt to a fine-tunedmodel with minimal samples. Specifically, we found that a model with a singlefine-tuning layer can be compromised using a single shot of adversarialsamples, while a model with more fine-tuning layers can be compromised usingonly a few shots.

Instance-level Trojan Attacks on Visual Question Answering via Adversarial Learning in Neuron Activation Space