New defense detects adversarial attacks on DNNs with 99.9% accuracy

Researchers have developed a novel method for detecting adversarial attacks on Deep Neural Networks (DNNs) used in autonomous driving systems. This approach leverages inconsistencies between multiple vision tasks, such as object detection and instance segmentation, to identify adversarial perturbations. The proposed defense achieves a high detection rate, with a ROC-AUC of 99.9% against PGD attacks on the BDD100k dataset, offering a more cost-efficient solution compared to existing methods. AI

IMPACT This research offers a more efficient and effective defense against adversarial attacks, crucial for the safety and reliability of AI systems in sensitive applications like autonomous driving.

RANK_REASON The cluster contains an academic paper detailing a new method for detecting adversarial attacks on AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• BDD100K
• Deep Neural Networks
• PGD attacks

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →