New research explores deep learning model coverage metrics and security

Researchers have conducted an empirical study to understand the relationships between deep learning model depth, configuration, and neural network coverage metrics. The study utilized LeNet, VGG, and ResNet architectures, along with models ranging from 5 to 54 layers, to analyze four coverage metrics: primary functionality, boundary, hierarchy, and structural coverage. Additionally, the research explored the connection between modified decision/condition coverage and dataset size, proposing three future research directions for enhancing DNN security testing. AI

IMPACT Provides insights into improving the security testing of deep learning models by analyzing coverage metrics.

RANK_REASON The cluster contains an academic paper detailing empirical research on deep learning models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• deep neural networks
• Hugging Face
• LeNet-5
• residual neural network
• Vgg Neural Network
• Wenkai Li

• paper
• safety

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