AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

MaskConnect: Connectivity Learning by Gradient Descent

Karim Ahmed, Lorenzo Torresani

ECCV 2018 · Architectures and efficiency › Neural architecture search · approach: Latency-aware architecture search · 54 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
add learned connectivity instead of fixed previous-block connectivityResNet-38 · CIFAR-100accuracy: 68.54 -> 70.40Improved
add learned connectivity instead of fixed previous-block connectivityResNet-110 · CIFAR-100accuracy: 71.21 -> 73.15Improved
learn connectivity rather than use random fixed connectivity with same fan-in (ablation)ResNet-38 · CIFAR-100accuracy: 62.67 -> 70.40Improved
learn connectivity rather than random fixed connectivity (ablation)ResNet-74 · CIFAR-100accuracy: 66.93 -> 72.81Improved
learn connectivity with K=4 instead of fixed full ResNeXt connectivityResNeXt {29,8,8} · CIFAR-100accuracy: 73.52 -> 75.89Improved
learn connectivity with K=4 instead of fixed full ResNeXt connectivityResNeXt {29,64,8} · CIFAR-100accuracy: 82.23 -> 84.05Improved
learn connectivity rather than random fixed with K=4 (ablation)ResNeXt {29,8,8} · CIFAR-100accuracy: 72.85 -> 75.89Improved
use fan-in K=1 learned connectivity (prunes unused blocks) instead of full connectivityResNeXt {29,64,8} · CIFAR-100accuracy and test params: 82.23 -> 82.31 with 34.4M -> 20.5M paramsImproved
add learned connectivity (K=10) to ResNet-50ResNet-50 · ImageNettop-1 accuracy: 76.15 -> 78.09Improved
add learned connectivity (K=C/2) to ResNeXtResNeXt {50,4,32} · ImageNettop-1 accuracy: 77.8 -> 79.1Improved
add learned connectivity (K=C/2) to ResNeXtResNeXt {101,4,64} · ImageNettop-1 accuracy: 79.6 -> 79.8Improved
add learned connectivity K=4 instead of fixed fullResNeXt {20,4,8} · Mini-ImageNettop-1 accuracy: 62.12 -> 66.09Improved

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