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

Developing lightweight Deep Convolutional Neural Networks (DCNNs) and VisionTransformers (ViTs) has become one of the focuses in vision research since thelow computational cost is essential for deploying vision models on edgedevices. Recently, researchers have explored highly computational efficientBinary Neural Networks (BNNs) by binarizing weights and activations ofFull-precision Neural Networks. However, the binarization process leads to anenormous accuracy gap between BNN and its full-precision version. One of theprimary reasons is that the Sign function with predefined or learned staticthresholds limits the representation capacity of binarized architectures sincesingle-threshold binarization fails to utilize activation distributions. Toovercome this issue, we introduce the statistics of channel information intoexplicit thresholds learning for the Sign Function dubbed DySign to generatevarious thresholds based on input distribution. Our DySign is a straightforwardmethod to reduce information loss and boost the representative capacity ofBNNs, which can be flexibly applied to both DCNNs and ViTs (i.e., DyBCNN andDyBinaryCCT) to achieve promising performance improvement. As shown in ourextensive experiments. For DCNNs, DyBCNNs based on two backbones (MobileNetV1and ResNet18) achieve 71.2% and 67.4% top1-accuracy on ImageNet dataset,outperforming baselines by a large margin (i.e., 1.8% and 1.5% respectively).For ViTs, DyBinaryCCT presents the superiority of the convolutional embeddinglayer in fully binarized ViTs and achieves 56.1% on the ImageNet dataset, whichis nearly 9% higher than the baseline.

Boosting Binary Neural Networks via Dynamic Thresholds Learning