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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep neural networks (DNNs) are increasingly being deployed insafety-critical systems such as personal healthcare devices and self-drivingcars. In such DNN-based systems, error resilience is a top priority sincefaults in DNN inference could lead to mispredictions and safety hazards. Forlatency-critical DNN inference on resource-constrained edge devices, it isnontrivial to apply conventional redundancy-based fault tolerance techniques.In this paper, we propose FitAct, a low-cost approach to enhance the errorresilience of DNNs by deploying fine-grained post-trainable activationfunctions. The main idea is to precisely bound the activation value of eachindividual neuron via neuron-wise bounded activation functions so that it couldprevent fault propagation in the network. To avoid complex DNN modelre-training, we propose to decouple the accuracy training and resiliencetraining and develop a lightweight post-training phase to learn theseactivation functions with precise bound values. Experimental results on widelyused DNN models such as AlexNet, VGG16, and ResNet50 demonstrate that FitActoutperforms state-of-the-art studies such as Clip-Act and Ranger in enhancingthe DNN error resilience for a wide range of fault rates while addingmanageable runtime and memory space overheads.

FitAct: Error Resilient Deep Neural Networks via Fine-Grained Post-Trainable Activation Functions