Neural network model interaction training method and device and storage medium

Abstract

The invention relates to a neural network model interaction training method and device. The method comprises the following steps: determining a main neural network participating in interaction training and at least one secondary neural network; determining an objective function participating in interactive training according to the distribution difference between the primary neural network and the secondary neural network; and training the primary neural network and the secondary neural network according to the objective function until the objective function value reaches a threshold value and tends to be stable, and obtaining the trained primary neural network. According to the neural network interaction training method, KL divergence is adopted to measure the prediction probability distribution difference of the primary network and the secondary network, primary network and secondary network interaction learning experience is used for guiding primary network learning, performance similar to or slightly higher than that of the secondary network is obtained, and the problems that convergence is slow when the primary network is independently trained, and learning efficiency is poor are solved. The method solves the problems that in the prior art, local optimum is likely to be trapped, and particularly under the training sample size limiting condition, the generalization of a network model is weak, and the detection rate is low.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and in particular relates to a neural network model interactive training method, device and storage medium. Background technique [0002] In recent years, deep learning neural networks have made remarkable achievements in the fields of computer vision, natural language processing, and intelligent speech recognition. Various application scenarios have become more and more mature, but in more complex environments, algorithm models often show Unstable prediction results and degraded application experience. The researchers found that the main reason for the above problems is that the model did not fully learn complex scene information during iterative training, resulting in inaccurate prediction results. [0003] In order to improve the performance of the network model, the current mainstream solution is to support iterative model training by collecting a large number of effective samples in v...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the accuracy of the neural network is not high when the training samples are limited, and the neural network is unstable due to the complex training environment during the training process, a neural network model interactive training method is provided in the first aspect of the present invention , comprising: determining a main neural network participating in interactive training, and at least one secondary neural network; determining an objective function participating in interactive training according to the distribution difference between the main neural network and the secondary neural network; training according to the objective function The main neural network and the secondary neural network until the objective function value reaches the threshold and tends to be stable, and the trained main neural network is obtained

Images