AI Paper Atlas

Research map / Architectures and efficiency / CNN design

PerforatedCNNs: Acceleration through Elimination of Redundant Convolutions

Mikhail Figurnov, Aizhan Ibraimova, Dmitry P Vetrov, Pushmeet Kohli

NeurIPS 2016 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 117 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune perforated network weights (ablation)PerforatedCNN VGG-16 CPU 4.0x · ImageNettop-5 error increase (%): +71.6 -> +5.5Improved
fine-tune perforated network weights (ablation)PerforatedCNN AlexNet GPU 4.1x · ImageNettop-5 error increase (%): +28.1 -> +6.2Improved
train from scratch/tune perforated network (ablation)PerforatedCNN NIN GPU 3.5x · CIFAR-10top-1 error increase (%): +37.4 -> +12.4Improved
combine perforation with linear asymmetric decompositionPerforation 4x · VGG-16 ImageNettuned error increase (%): +5.5 -> +1.6Improved
combine perforation with linear asymmetric decompositionAsymmetric decomposition 4x · VGG-16 ImageNettuned error increase (%): +2.0 -> +1.6Improved
combine perforation with asymmetric decompositionPerforation · VGG-16 ImageNetmemory reduction: 2.5x -> 1.4xImproved
reduce conv2 kernel to 3x3 with impact maskAlexNet conv2 impact mask · ImageNetCPU speedup at about 1% top-5 error increase: 4.2x -> 9.1xImproved
use impact mask vs grid maskPerforation masks · AlexNet conv2 without fine-tuningspeedup-error tradeoffImproved
replace missing values with zero instead of nearest neighbor (ablation)Perforation interpolation · AlexNet conv3errorHurt
use barycentric interpolation instead of nearest neighbor (ablation)Perforation interpolation · AlexNet conv2/conv3errorMixed
accelerate layers followed by max-pooling vs followed by convolutionPerforation · AlexNet conv2 vs conv3speedup-error tradeoffImproved
use uniform mask at high speedupsPerforation uniform mask · AlexNeterror standard deviationHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy