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

Recent works on neural network pruning advocate that reducing the depth ofthe network is more effective in reducing run-time memory usage andaccelerating inference latency than reducing the width of the network throughchannel pruning. In this regard, some recent works propose depth compressionalgorithms that merge convolution layers. However, the existing algorithms havea constricted search space and rely on human-engineered heuristics. In thispaper, we propose a novel depth compression algorithm which targets generalconvolution operations. We propose a subset selection problem that replacesinefficient activation layers with identity functions and optimally mergesconsecutive convolution operations into shallow equivalent convolutionoperations for efficient end-to-end inference latency. Since the proposedsubset selection problem is NP-hard, we formulate a surrogate optimizationproblem that can be solved exactly via two-stage dynamic programming within afew seconds. We evaluate our methods and baselines by TensorRT for a fairinference latency comparison. Our method outperforms the baseline method withhigher accuracy and faster inference speed in MobileNetV2 on the ImageNetdataset. Specifically, we achieve $1.41\times$ speed-up with $0.11$\%p accuracygain in MobileNetV2-1.0 on the ImageNet.

Efficient Latency-Aware CNN Depth Compression via Two-Stage Dynamic Programming