CheckOne: Lightweight Fault Detection and Mitigation for Vision Transformers

2026-08-03

AI Summary

This work addresses the vulnerability of Vision Transformers (ViTs) to hardware faults in safety-critical applications, where existing fault-tolerance techniques incur prohibitive computational overhead and deployment complexity. To overcome these limitations, the authors propose CheckOne, a lightweight, symmetry-aware protection mechanism tailored for ViTs. By integrating algorithm-based fault tolerance (ABFT) with architecture-aware fault detection and mitigation strategies, CheckOne enables efficient error resilience across all network layers. The approach substantially reduces fault-tolerance overhead while achieving up to 26× higher critical fault mitigation capability across various ViT models compared to baseline methods, with an average performance 3.8× better than conventional ABFT schemes—effectively balancing reliability and efficiency.

Abstract

The wide adoption of Vision Transformers (ViTs) in safety-critical applications raises reliability concerns related to hardware faults. Algorithm-Based Fault Tolerance (ABFT) methods have emerged as lightweight and symmetric protection mechanisms for DNNs. However, they are particularly challenging for ViTs due to their significant computational requirements. This work comprehensively evaluates the reliability of ViTs, emphasizing the need for symmetric protection in their layers. Furthermore, we present CheckOne, a novel, cost-effective method for fault detection and mitigation in ViTs that significantly reduces the computational cost compared to conventional ABFT. Through extensive experiments with multiple ViTs, CheckOne mitigates critical faults by up to $26\times$ and achieves an average 3.8x higher performance than ABFT in ViTs.

Problem

• Vision Transformers
• fault detection
• hardware faults
• Algorithm-Based Fault Tolerance
• reliability

Innovation

• Vision Transformers
• fault detection
• Algorithm-Based Fault Tolerance
• lightweight protection
• reliability