Training method and device of neural network model for sample classification

Abstract

The embodiment of the invention provides a training method and device of a neural network model for sample classification. The method comprises the steps of obtaining training samples in a training sample set, wherein the training samples have sample identifiers and sample category labels; inputting the training sample into a feature extraction model to obtain a feature representation vector; inputting the feature representation vector into a discriminator model to obtain an identification mark; determining a first prediction loss according to the identification identifier and the sample identifier, and performing first training on the discriminator model and the feature extraction model by taking minimization of the first prediction loss as a target; inputting the feature representation vector into a classifier model to obtain an identification category; and determining a second prediction loss according to the identification category and the sample category label, determining a thirdprediction loss according to the negative correlation with the first prediction loss and the positive correlation with the second prediction loss, and carrying out second training on the classifier model and the feature extraction model by taking minimization of the third prediction loss as a target. And the generalization of the model can be improved.

Description

technical field [0001] One or more embodiments of this specification relate to the computer field, and in particular to a training method and device for a neural network model for sample classification. Background technique [0002] In deep learning, a larger neural network model is usually set up to fit the training data so that the neural network model can be used for sample classification. However, due to too many parameters, the complexity of the neural network model is too high. In most cases, the neural network model will be overfitted, and the generalization of the trained neural network model will not be the best. [0003] Therefore, it is hoped that there will be an improved solution, which can prevent the neural network model from overfitting and improve the generalization of the trained neural network model during the training process of the neural network model. Contents of the invention [0004] One or more embodiments of this specification describe a trainin...

AI Technical Summary

Problems solved by technology

However, due to too many parameters, the complexity of the neural network model is too high. In most cases, the neural network model will be overfitted, and the generalization of the trained neural network model will not be the best.

Images