NASNet

E899014

NASNet variant convolutional neural network architecture family deep learning model image classification model neural architecture search result

NASNet is a family of convolutional neural network architectures automatically discovered via neural architecture search, known for achieving state-of-the-art performance on image classification benchmarks.

Jump to: Surface forms Statements Referenced by

Observed surface forms (5)

Surface form	Occurrences
NASNet-A	0
NASNet-A-Mobile	0
NASNet-B	0
NASNet-C	0
NASNet-Large	0

Statements (51)

Predicate	Object
instanceOf	NASNet variant ⓘ NASNet variant ⓘ NASNet variant ⓘ NASNet variant ⓘ NASNet variant ⓘ convolutional neural network architecture family ⓘ deep learning model ⓘ image classification model ⓘ neural architecture search result ⓘ
achievedStateOfTheArtOn	CIFAR-10 test set ⓘ ImageNet validation set NERFINISHED ⓘ
basedOn	neural architecture search ⓘ
controllerType	recurrent neural network controller ⓘ
developedBy	Barret Zoph NERFINISHED ⓘ Google Brain NERFINISHED ⓘ Jonathon Shlens NERFINISHED ⓘ Quoc V. Le NERFINISHED ⓘ Vijay Vasudevan NERFINISHED ⓘ
field	automated machine learning ⓘ computer vision ⓘ
hasDesignPrinciple	cell-based architecture search ⓘ search on small dataset then transfer to large dataset ⓘ
hasLicense	Apache License 2.0 (for TensorFlow implementation) NERFINISHED ⓘ
hasVariant	NASNet-A NERFINISHED ⓘ NASNet-A-Mobile NERFINISHED ⓘ NASNet-B NERFINISHED ⓘ NASNet-C NERFINISHED ⓘ NASNet-Large NERFINISHED ⓘ
implementedIn	TensorFlow NERFINISHED ⓘ
influenced	AmoebaNet NERFINISHED ⓘ EfficientNet NERFINISHED ⓘ
introducedInPaper	Learning Transferable Architectures for Scalable Image Recognition NERFINISHED ⓘ
introducedYear	2017 ⓘ
optimizedFor	accuracy ⓘ computational efficiency ⓘ computational efficiency ⓘ high accuracy ⓘ mobile and embedded devices ⓘ
paperArchiveId	arXiv:1707.07012 ⓘ
searchedOnDataset	CIFAR-10 NERFINISHED ⓘ
searchMethod	reinforcement learning controller ⓘ
searchSpace	convolutional cell structures ⓘ
task	image classification ⓘ
top1AccuracyOnImageNetApprox	82.7% ⓘ
top5AccuracyOnImageNetApprox	96.2% ⓘ
transferredToDataset	ImageNet NERFINISHED ⓘ
uses	convolutional layers ⓘ normal cell ⓘ reduction cell ⓘ
usesRegularization	batch normalization ⓘ dropout ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Neural Architecture Search → notableMethod → NASNet ⓘ