Method, device and system for training neural network and a storage medium

Abstract

The invention provides a method, device and system for training a neural network and a storage medium. The method comprises the steps of performing a round of training on a to-be-trained first neuralnetwork based on training data; based on the respective state of each piece of training data in the training data after the current round of training, adopting a second neural network to generate therespective weight of each piece of training data used for performing the next round of training on the first neural network; carrying out the next round of training of the first neural network based on each piece of training data with the weight, and returning to the step of generating the weight so as to generate the weight of each piece of training data for carrying out the next round of training on the first neural network, and carrying out iteration in this way until a preset training round number is reached or a loss function of the first neural network is converged. The training convergence process of the neural network can be accelerated, and the network precision of the trained neural network can be improved.

Description

technical field [0001] The present invention relates to the technical field of deep learning, and more specifically relates to a method, device, system and storage medium for training a neural network. Background technique [0002] Deep learning (Deep Learning, DL) is one of the technology and research fields of machine learning. By establishing artificial neural networks (Artifitial Neural Networks, ANNs) with hierarchical structure, artificial intelligence is realized in computing systems. With the wide application of deep learning, the quality of the training data used in training the neural network also plays a decisive role in the training process of the network model. Therefore, how to select training data that is friendly to the model in the current training state is also very important. [0003] At present, the following methods are mainly used to select samples during neural network training: (1) online hard example mining (ohem for short) method, that is, to selec...

AI Technical Summary

Problems solved by technology

However, these methods cannot avoid artificial rules, ignoring some potential influencing factors

Images