Papers › GreedyNAS: Towards Fast One-Shot NAS with Greedy Supernet

GreedyNAS: Towards Fast One-Shot NAS with Greedy Supernet

25 Mar 2020CVPR 2020 6arXiv:2003.11236archive 2025-07-28

Shan You, Tao Huang, Mingmin Yang, Fei Wang, Chen Qian, Chang-Shui Zhang

Training a supernet matters for one-shot neural architecture search (NAS) methods since it serves as a basic performance estimator for different architectures (paths). Current methods mainly hold the assumption that a supernet should give a reasonable ranking over all paths. They thus treat all paths equally, and spare much effort to train paths. However, it is harsh for a single supernet to evaluate accurately on such a huge-scale search space (e.g., 7²¹). In this paper, instead of covering all paths, we ease the burden of supernet by encouraging it to focus more on evaluation of those potentially-good ones, which are identified using a surrogate portion of validation data. Concretely, during training, we propose a multi-path sampling strategy with rejection, and greedily filter the weak paths. The training efficiency is thus boosted since the training space has been greedily shrunk from all paths to those potentially-good ones. Moreover, we further adopt an exploration and exploitation policy by introducing an empirical candidate path pool. Our proposed method GreedyNAS is easy-to-follow, and experimental results on ImageNet dataset indicate that it can achieve better Top-1 accuracy under same search space and FLOPs or latency level, but with only ∼60\% of supernet training cost. By searching on a larger space, our GreedyNAS can also obtain new state-of-the-art architectures.

PaperPDFConference PDF

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

AllImage ClassificationNeural Architecture Search

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Classification ImageNet GreedyNAS-A GFLOPs 0.366 #887 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-A Number of params 6.5M #887 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-A Top 1 Accuracy 77.1% #887 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-B GFLOPs 0.324 #898 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-B Number of params 5.2M #898 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-B Top 1 Accuracy 76.8% #898 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-C GFLOPs 0.284 #922 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-C Number of params 4.7M #922 of 1060 Archive leaderboard report
Image Classification ImageNet GreedyNAS-C Top 1 Accuracy 76.2% #922 of 1060 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-A Accuracy 77.1 #72 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-A Params 6.5M #72 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-A Top-1 Error Rate 22.9 #72 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-B Accuracy 76.8 #79 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-B Params 5.2M #79 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-B Top-1 Error Rate 23.2 #79 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-C Accuracy 76.2 #89 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-C Params 4.7M #89 of 135 Archive leaderboard report
Neural Architecture Search ImageNet GreedyNAS-C Top-1 Error Rate 23.8 #89 of 135 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: GreedyNAS, GreedyNAS-B, GreedyNAS-C

1x1 ConvolutionAverage PoolingBatch NormalizationConvolutionCosine AnnealingDense ConnectionsDepthwise ConvolutionDepthwise Separable ConvolutionGreedyNASGreedyNAS-AGreedyNAS-BGreedyNAS-CInverted Residual BlockLinear Warmup With Linear DecayNesterov Accelerated GradientPointwise ConvolutionRMSPropReLUSigmoid ActivationSqueeze-and-Excitation Block

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections