NEURAL NETWORK ARCHITECTURE SEARCH METHOD AND APPARATUS, DEVICE AND MEDIUM

摘要

Provided by the present application is a neural network architecture search method, which is applied to a search system. The search system comprises a generator and a searcher. The method comprises that: the generator generates a plurality of neural network architectures according to a search space; the searcher acquires evaluation index values of a plurality of sub-models on first hardware that are obtained according to the plurality of neural network architectures; and according to the neural network architectures corresponding to the plurality of sub-models and the evaluation index values of the plurality of sub-models on the first hardware, the searcher determines a first target neural network architecture that meets a preset condition. In the foregoing manner, the decoupling of different initial sub-model training processes and the decoupling of the neural network architecture search process and the initial sub-model training process are achieved, the search time is shortened, and the search efficiency is improved.