Characterizing Deep Neural Networks Neutrons-Induced Error Model
Résumé
We characterize the fault models for Deep Neural Networks (DNNs) in GPUs exposed to neutron. We observe tolerable and critical errors, and show that ECC is not effective in reducing critical errors.
Domaines
Informatique [cs]| Origine | Fichiers produits par l'(les) auteur(s) |
|---|
