Progressive Neural Architecture Search

  • 2017-12-02 06:23:16
  • Chenxi Liu, Barret Zoph, Jonathon Shlens, Wei Hua, Li-Jia Li, Li Fei-Fei, Alan Yuille, Jonathan Huang, Kevin Murphy
  • 136


We propose a method for learning CNN structures that is more efficient thanprevious approaches: instead of using reinforcement learning (RL) or geneticalgorithms (GA), we use a sequential model-based optimization (SMBO) strategy,in which we search for architectures in order of increasing complexity, whilesimultaneously learning a surrogate function to guide the search, similar to A*search. On the CIFAR-10 dataset, our method finds a CNN structure with the sameclassification accuracy (3.41% error rate) as the RL method of Zoph et al.(2017), but 2 times faster (in terms of number of models evaluated). It alsooutperforms the GA method of Liu et al. (2017), which finds a model with worseperformance (3.63% error rate), and takes 5 times longer. Finally we show thatthe model we learned on CIFAR also works well at the task of ImageNetclassification. In particular, we match the state-of-the-art performance of82.9% top-1 and 96.1% top-5 accuracy.


Introduction (beta)



Conclusion (beta)