Pruning, adversarial training, and hardware faults interact to affect DNN reliability

A new research paper investigates the combined impact of model pruning, adversarial training, and hardware faults on the reliability of deep neural networks. The study found that while adversarial training enhances robustness against input perturbations, it also increases sensitivity to hardware-induced weight faults. Contrary to expectations, pruning did not significantly degrade fault tolerance or adversarial robustness. The findings underscore the importance of considering both adversarial robustness and hardware reliability concurrently when deploying models on resource-constrained devices. AI

IMPACT Highlights the need for joint consideration of adversarial robustness and hardware reliability in model deployment.

RANK_REASON Research paper published on arXiv detailing empirical investigation of model fault tolerance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• convolutional neural network
• Deep Neural Networks
• MNIST database
• Neuromorphic Hardware Architecture Using the Neural Engineering Framework for Pattern Recognition

• paper
• safety

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