AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Zero-Shot Neural Architecture Search: Challenges, Solutions, and Opportunities

Guihong Li, Duc Hoang, Kartikeya Bhardwaj, Ming Lin, Zhangyang Wang, Radu Mărculescu

TPAMI 2024 · Architectures and efficiency › Neural architecture search · approach: One-shot and few-shot supernets · 66 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
use #Params instead of Zen-score to pick architecturezero-shot NAS · NASBench-201 ImageNet16-120best accuracy (%): 40.77 -> 41.44Improved
use #Params instead of Synflowzero-shot NAS · NASBench-201 CIFAR-100best accuracy (%): 62.22 -> 71.11Improved
use Zen-score instead of #Paramszero-shot NAS · NATS-Bench ImageNet16-120best accuracy (%): 44.73 -> 46.27Improved
restrict correlation to top 5% architecturesexisting proxies · NASBench-201 and NATS-Benchrank correlationHurt
use Jacob_cov or Zen-score instead of #Params on non-classification taskszero-shot proxies · TransNAS-Bench-101correlationImproved
use #Params instead of Fisher for best architecturezero-shot NAS · TransNAS-Bench-101 autoencodingSSIM: 0.49 -> 0.46Hurt
use one-shot NAS instead of zero-shot #ParamsNAS · ImageNet-1K ProxylessNAS spacetop-1 accuracy (%): 72.87 -> 74.39Improved
use NTK_Cond instead of #Paramszero-shot NAS · ImageNet-1K ProxylessNAS spacetop-1 accuracy (%): 72.87 -> 73.63Improved
use #Params instead of NTK_Condzero-shot NAS · ImageNet-1K ProxylessNAS spacesearch cost (GPU hours): 37 -> 0.03Improved
use zero-shot #Params instead of one-shot NASNAS · ImageNet-1K ProxylessNAS spacesearch cost (GPU hours): 200 -> 0.03Improved
use #FLOPs instead of #Paramszero-shot NAS · COCO detectionmAP: 0.26 -> 0.28Improved
use proxies within ResNet and Wide-ResNet familiesSNIP, Zen-score, #Params, #FLOPs, NN-Mass · ImageNet-1K ResNet familySpearman rho: rho > 0.9Improved

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