English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

With the increasing deployment of deep neural networks (DNNs) in terrestrialand aerospace safety-critical applications, system reliability has emerged as aco-equal design metric alongside computational efficiency. Algorithm-basedfault tolerance (ABFT) mechanisms, characterized by architecture-agnostic andcost-effectiveness, have become a promising solution for reliabilityenhancement. However, conventional ABFT approaches rely on rigorousverification mechanisms where even minor computational deviations trigger errorrecovery processes, which not only disregards the intrinsic fault tolerancecharacteristics of DNN models but also incurs redundant fault toleranceprocessing overhead. To address these limitations, we propose an ApproximateABFT framework (ApproxABFT) that innovatively introduces adaptive errortolerance thresholds to enable selective fault recovery, activating errorcorrection modules exclusively when computational deviations exceed predefinedthresholds. This approach effectively mitigating overreaction to non-criticalcomputational errors. Furthermore, a dynamic block granularity optimizationalgorithm is implemented to achieve inter-layer error sensitivity balancing.Experimental evaluations demonstrate that the proposed ApproxABFT achieves a43.39% average reduction in redundant computing overhead compared to previousaccurate ABFT, while simultaneously enhancing the tolerable soft error rate byan order of magnitude.

ApproxABFT: Approximate Algorithm-Based Fault Tolerance for Neural Network Processing