MaxiMals: A Low-cost Hardening Technique for Large Vision Transformers - Archive ouverte HAL
Communication Dans Un Congrès Année : 2024

MaxiMals: A Low-cost Hardening Technique for Large Vision Transformers

Résumé

We propose MaxiMals, an experimentally tuned lowcost mitigation solution to reduce the impact of transient faults on Vision Transformers. MaxiMals can correct 73.06% of critical failures on average, with overheads as low as 0.25%.
Fichier principal
Vignette du fichier
radecs_2024_maximals.pdf (613.81 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04736704 , version 1 (15-10-2024)

Licence

Identifiants

  • HAL Id : hal-04736704 , version 1

Citer

Lucas Roquet, Fernando Fernandes dos Santos, Paolo Rech, Marcello Traiola, Olivier Sentieys, et al.. MaxiMals: A Low-cost Hardening Technique for Large Vision Transformers. RADECS 2024 - Conference is an annual on RADiation Effects on Components and Systems, RADECS Association, Sep 2024, Maspalomas, Canary Islands, Spain. pp.1-5. ⟨hal-04736704⟩
26 Consultations
27 Téléchargements

Partager

More