AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

AttentiveNAS: Improving Neural Architecture Search via Attentive Sampling

Dilin Wang, Meng Li, Chengyue Gong, Vikas Chandra

CVPR 2021 · Architectures and efficiency › Neural architecture search · approach: Efficient NAS search strategies · 91 citations (OpenAlex)

Check 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
Use WorstUp-50 accuracy-predictor sampling instead of uniform sampling (ablation)Two-stage NAS · ImageNet, 200±10 MFLOPsTop-1 accuracy: around 0.3%Improved
Focus training on the worst Pareto architectures rather than uniform sampling (ablation)Two-stage NAS · ImageNet, low-FLOPs regimesMinimum and first-quartile accuracyImproved
Use WorstUp-1M accuracy-predictor sampling rather than BestUp-1M (ablation)Two-stage NAS · ImageNetTop-1 accuracy across FLOPs regimesImproved
Use BestUp-50 accuracy-predictor sampling rather than uniform sampling (ablation)Two-stage NAS · ImageNet, medium-FLOPs regimes above 400 MFLOPsMaximum top-1 accuracyImproved
Use BestUp-3 minibatch-loss sampling rather than uniform sampling (ablation)Two-stage NAS · ImageNetMaximum top-1 accuracyImproved
Use WorstUp-3 minibatch-loss sampling rather than uniform sampling (ablation)Two-stage NAS · ImageNetMaximum top-1 accuracyImproved
Balance exploration and exploitation with BestUp-3 minibatch-loss sampling rather than accuracy-predictor-based BestUp (ablation)Two-stage NAS · ImageNetMaximum top-1 accuracyImproved
Increase the number of sampled candidate architectures from BestUp-50 to offline BestUp-1M (ablation)BestUp accuracy-predictor sampling · ImageNetMaximum top-1 accuracyMixed
Use accuracy predictors trained early to rank candidate subnetworks (ablation)AttentiveNAS performance estimation · ImageNet subnetworksKendall's τ: 0.89; 0.87; 0.88Improved

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