AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Genetic CNN

Lingxi Xie, Alan Yuille

ICCV 2017 · Architectures and efficiency › Neural architecture search · approach: Latency-aware architecture search · 930 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
run 50 generations of genetic search instead of the initial random populationGeNet · CIFAR10 testerror rate (%): error 24.04% -> 22.81%Improved
replace random initialization with all-zero initialization (ablation)GeNet · CIFAR10 testaccuracy (%)No clear effect
use genetic search instead of random search with same budget of 1020 individuals (ablation)GeNet · CIFAR10 testaccuracy (%): 76.94 -> 77.19Improved
use the searched structure instead of the densely connected all-ones string (ablation)GeNet · CIFAR10 testaccuracy (%): 76.84 -> 77.19Improved
run longer genetic search (G-00 to G-50), transferred to CIFAR10 (ablation)GeNet · CIFAR10 error (Table 2)error rate (%): 8.18 -> 7.19Improved
run longer genetic search (G-00 to G-50), transferred to CIFAR100 (ablation)GeNet · CIFAR100 error (Table 2)error rate (%): 31.46 -> 29.03Improved
run longer genetic search (G-00 to G-50), transferred to SVHN (ablation)GeNet · SVHN error (Table 2)error rate (%): 2.25 -> 1.99Improved
replace VGGNet-16 chain stages with GeNet #2 structureVGGNet-16 · ILSVRC2012 val top-1top-1 error (%): 28.5 -> 27.87Improved
replace VGGNet-16 chain stages with GeNet #2 structureVGGNet-16 · ILSVRC2012 val top-5top-5 error (%): 9.9 -> 9.74Improved
replace VGGNet-16 chain stages with GeNet #1 structureVGGNet-16 · ILSVRC2012 val top-5top-5 error (%): 9.9 -> 9.95No clear effect
start genetic search from the 40-layer wide residual networkWRN-40 · CIFAR10 errorerror rate (%): 5.54 -> 5.39Improved
replace VGGNet-16 features with GeNet #2 featuresVGGNet-16 · Caltech256accuracy (%): 82.69 -> 83.78Improved

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