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

Compressing convolutional neural networks (CNNs) has received ever-increasingresearch focus. However, most existing CNN compression methods do not interprettheir inherent structures to distinguish the implicit redundancy. In thispaper, we investigate the problem of CNN compression from a novel interpretableperspective. The relationship between the input feature maps and 2D kernels isrevealed in a theoretical framework, based on which a kernel sparsity andentropy (KSE) indicator is proposed to quantitate the feature map importance ina feature-agnostic manner to guide model compression. Kernel clustering isfurther conducted based on the KSE indicator to accomplish high-precision CNNcompression. KSE is capable of simultaneously compressing each layer in anefficient way, which is significantly faster compared to previous data-drivenfeature map pruning methods. We comprehensively evaluate the compression andspeedup of the proposed method on CIFAR-10, SVHN and ImageNet 2012. Our methoddemonstrates superior performance gains over previous ones. In particular, itachieves 4.7 \times FLOPs reduction and 2.9 \times compression on ResNet-50with only a Top-5 accuracy drop of 0.35% on ImageNet 2012, which significantlyoutperforms state-of-the-art methods.

Exploiting Kernel Sparsity and Entropy for Interpretable CNN Compression