TreeFI methodology slashes DNN fault injection costs by up to 72x

Researchers have developed TreeFI, a novel value-aware statistical fault injection methodology designed to improve the reliability evaluation of deep neural networks. This approach specifically targets single-bit faults in FP32 activations and weights, partitioning value distributions into intervals with similar fault behaviors using regression trees. By allocating injections based on relevance for failure-rate estimation, TreeFI significantly reduces the required injection budget, achieving reductions of up to 72.1x compared to existing methods. The methodology has been validated on CNN and Transformer models using CIFAR-10 and ImageNet datasets, demonstrating more accurate estimates with fewer injections than state-of-the-art baselines. AI

IMPACT Reduces the computational cost of evaluating DNN reliability, potentially accelerating the deployment of more robust AI systems.

RANK_REASON Academic paper detailing a new methodology for DNN reliability evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• CIFAR-10
• CNN
• Deep Neural Networks
• ImageNet
• ResNet8
• FP32
• Transformer++

• paper
• infra

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