Skip to main content

Resultados relacionados

Resultados totales (Incluyendo duplicados): 1
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Exploiting neural networks Bit-level redundancy to mitigate the impact of faults at inference

  • Catalán-Gallach, Izan|||0000-0001-7993-433X
  • Flich Cardo, José
  • Hernández Luz, Carles|||0000-0003-1192-6223
[EN] Neural networks are widely used in critical environments such as healthcare, autonomous vehicles, or video surveillance. To ensure the safety of the systems that rely on their functionality, it is essential to validate their correct behaviour in the presence of faults. This paper studies the behaviour of state-of-the-art neural network models with fault injection in their weights. For this purpose, we analyse the sensitivity of these models and identify the impact of bit flips on their accuracy. To mitigate the effects of faults, we introduce two mechanisms that leverage bit-level redundancy for protection. The first mechanism, Fixed Protection, safeguards consecutive sets of bits, while the second, Variable Protection, targets non-consecutive bits. Our findings demonstrate that, on average, random bit flip faults cause the accuracy of the original models to drop by 1.3% to over 3%. However, with our protection mechanisms in place, accuracy reductions are significantly minimised, ranging from only 0.0001% to 0.4%., This article has been made possible thanks to the Grant "BECA FPU PROGRAMA PROPIO DE LA UNIVERSITAT POLITECNICA DE VALENCIA - SUBPROGRAMA 1 (PAID-01-20)" No. 20210037. Carles Hernandez is partially supported by the Spanish Ministry of Science, Innovation and Universities under "Ramon y Cajal", fellowship No. RYC2020-030685-I. In addition, this work has received funding from the Valencian government project "IFAC: Implementing Fault-Tolerant Autonomous Computers (CISEJI/2022/30)", and from the Chips Joint Undertaking project EdgeAITrust no. 101139892 and the Spanish Agencia Estatal de Investigacion PCI2024-155056-2 grant. Funding for open access charge: CRUE-Universitat Politecnica de Valencia.