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

Distributed deep learning has recently been attracting more attention inremote sensing (RS) applications due to the challenges posed by the increasedamount of open data that are produced daily by Earth observation programs.However, the high communication costs of sending model updates among multiplenodes are a significant bottleneck for scalable distributed learning. Gradientsparsification has been validated as an effective gradient compression (GC)technique for reducing communication costs and thus accelerating the trainingspeed. Existing state-of-the-art gradient sparsification methods are mostlybased on the "larger-absolute-more-important" criterion, ignoring theimportance of small gradients, which is generally observed to affect theperformance. Inspired by informative representation of manifold structures fromneighborhood information, we propose a simple yet effective dynamic gradientcompression scheme leveraging neighborhood statistics indicator for RS imageinterpretation, termed RS-DGC. We first enhance the interdependence betweengradients by introducing the gradient neighborhood to reduce the effect ofrandom noise. The key component of RS-DGC is a Neighborhood StatisticalIndicator (NSI), which can quantify the importance of gradients within aspecified neighborhood on each node to sparsify the local gradients beforegradient transmission in each iteration. Further, a layer-wise dynamiccompression scheme is proposed to track the importance changes of each layer inreal time. Extensive downstream tasks validate the superiority of our method interms of intelligent interpretation of RS images. For example, we achieve anaccuracy improvement of 0.51% with more than 50 times communication compressionon the NWPU-RESISC45 dataset using VGG-19 network.

RS-DGC: Exploring Neighborhood Statistics for Dynamic Gradient Compression on Remote Sensing Image Interpretation