This work assesses the reliability of a hyperspectral image classifier for edge devices under transient faults by using a fine-grain strategy based on the Hardware Injection Through Program Transformation (HITPT) technique. The results identified the most vulnerable software parts and the corruption effects due to hardware faults (from 5.1% to 100.0% of accuracy drop). Then, the results supported the adoption of a selective-hardening software mechanism (based on the Duplication with Comparison strategy) to effectively mitigate the most critical effects under limited costs.
Analysis and Mitigation of Soft-errors in GPU-accelerated Hyperspectral Image Classifiers / Abed, Sergiu-Mohamed; Guerrero-Balaguera, Juan-David; Condia, Josie E. Rodriguez; De Lucia, Gianluca; Lapegna, Marco; Reorda, Matteo Sonza. - (2025), pp. 131-134. [10.1109/ddecs63720.2025.11006812]
Analysis and Mitigation of Soft-errors in GPU-accelerated Hyperspectral Image Classifiers
De Lucia, Gianluca;Lapegna, Marco;
2025
Abstract
This work assesses the reliability of a hyperspectral image classifier for edge devices under transient faults by using a fine-grain strategy based on the Hardware Injection Through Program Transformation (HITPT) technique. The results identified the most vulnerable software parts and the corruption effects due to hardware faults (from 5.1% to 100.0% of accuracy drop). Then, the results supported the adoption of a selective-hardening software mechanism (based on the Duplication with Comparison strategy) to effectively mitigate the most critical effects under limited costs.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
