Neural network structure model search method and device and storage medium

Abstract

The invention relates to a neural network structure model search method and device and a storage medium. The method comprises: generating an initial-generation network model structure population basedon the multi-objective optimization hyper-parameters; selecting and crossing the current generation network model structure population; generating partial network model structures based on reinforcement learning variation, and performing random variation on the selected and crossed network model structures to generate other partial network model structures; generating a new population of a network model structure based on the network model structure generated by reinforcement learning variation and the network model structure generated by random variation; searching a next-generation networkmodel structure population based on the current-generation network model structure population and the network model structure new population; and taking the next generation of network model structurepopulation as the current generation of network model structure population, repeatedly executing the process until the optimal multi-objective optimization state is achieved, and selecting neural network structure models suitable for different scenes from the final generation of network model structure population.

Description

technical field [0001] The present disclosure relates to the technical field of deep learning, and in particular to a neural network structure model search method, device and storage medium. Background technique [0002] The deep learning neural network achieves end-to-end feature extraction, which is a huge improvement compared to manual feature extraction. At the same time, the artificially designed neural network architecture increases the complexity of the neural network. [0003] With the development of technology, neural network architecture search (Neural Architecture Search, NAS) realizes the use of neural network to design neural network, which represents the future direction of machine learning. In NAS technology, an evolutionary algorithm or a reinforcement learning algorithm is used as a search strategy to search for a neural network structure model. Contents of the invention [0004] In order to overcome the problems existing in related technologies, the pres...

AI Technical Summary

Problems solved by technology

[0002] The deep learning neural network achieves end-to-end feature extraction, which is a huge improvement compared to manual feature extraction. At the same time, the artificially designed neural network architecture increases the complexity of the neural network.

Images