Learning Transferable Architectures for Scalable Image Recognition

Developingneuralnetworkimageclassificationmodelsoftenrequiressignificantarchitectureengineering.Inthispaper,westudyamethodtolearnthemodelarchitecturesdirectlyonthedatasetofinterest.Asthisapproachisexpensivewhenthedatasetislarge,weproposetosearchforanarchitecturalbuildingblockonasmall