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

Network pruning has become the de facto tool to accelerate deep neuralnetworks for mobile and edge applications. Recently, feature-map discriminantbased channel pruning has shown promising results, as it aligns well with theCNN objective of differentiating multiple classes and offers betterinterpretability of the pruning decision. However, existing discriminant-basedmethods are challenged by computation inefficiency, as there is a lack oftheoretical guidance on quantifying the feature-map discriminant power. In thispaper, we present a new mathematical formulation to accurately and efficientlyquantify the feature-map discriminativeness, which gives rise to a novelcriterion,Discriminant Information(DI). We analyze the theoretical property ofDI, specifically the non-decreasing property, that makes DI a valid selectioncriterion. DI-based pruning removes channels with minimum influence to DIvalue, as they contain little information regarding to the discriminant power.The versatility of DI criterion also enables an intra-layer mixed precisionquantization to further compress the network. Moreover, we propose a DI-basedgreedy pruning algorithm and structure distillation technique to automaticallydecide the pruned structure that satisfies certain resource budget, which is acommon requirement in reality. Extensive experiments demonstratetheeffectiveness of our method: our pruned ResNet50 on ImageNet achieves 44% FLOPsreduction without any Top-1 accuracy loss compared to unpruned model

A Feature-map Discriminant Perspective for Pruning Deep Neural Networks