ApproxABFT: Approximate Algorithm-Based Fault Tolerance for Vision Transformers. (arXiv:2302.10469v1 [cs.CR])

Vision Transformers (ViTs) with outstanding performance becomes a popular
backbone of deep learning models for the main-stream vision tasks including
classification, object detection, and segmentation. Other than the performance,
reliability is also a critical metric for the adoption of ViTs in
safety-critical applications such as autonomous driving and robotics. With the
observation that the major computing blocks in ViTs such as multi-head
attention and feed forward are usually performed with general matrix
multiplication (GEMM), we propose to adopt a classical …

adoption algorithm applications attention autonomous autonomous driving classification computing critical deep learning detection driving forward head main major object performance popular reliability robotics safety safety-critical segmentation stream tolerance transformers