TreeFI: Value-Aware Statistical Fault Injection for Deep Neural Networks
TreeFI, a value-aware statistical fault-injection methodology for FP32 single-bit faults in DNN activations and weights, is proposed and validated, providing more accurate estimates than state-of-the-art statistical FI baselines under the same campaign setting.
Noam Bires, Marcello Traiola, Angeliki Kritikakou et al.
· 0 citations