Classification model training method and device, computing equipment and storage medium

Abstract

The embodiment of the invention provides a classification model training method and device, computing equipment and a storage medium. The method comprises the steps: inputting a training sample into a classification model, and obtaining the prediction probability distribution of the training sample belonging to each class; and calculating target loss and a penalty term, wherein the penalty term is used for indicating the dispersion degree of the negative class in the prediction probability distribution. The sum of the target loss and the penalty term is recorded as total loss, and classification model parameters are updated according to the total loss. In traditional model training, due to the limitation of a loss function, the accuracy of the model on negative class prediction is ignored. The dispersion degree of negative class distribution is used as the penalty term to be introduced into the model loss to construct a novel loss function, the performance of the original loss function is improved, and the prediction capability of the model is improved.

Description

technical field [0001] The present application relates to the field of deep learning, and specifically relates to a classification model training method, device, computing device and storage medium. Background technique [0002] The purpose of multi-classification tasks is to assign appropriate class labels to an input data. In the multi-classification task, there is only one category label of the data, and the category label probability is predicted through the classification model, and the label with the highest probability is used as the category of the data. [0003] In the training of the classification model, the loss function is used to update the parameters. The loss function commonly used in multi-classification tasks is cross-entropy loss. However, the cross-entropy loss function only cares about the accuracy of the prediction probability of positive labels, which leads to low prediction accuracy of the trained model in actual use and cannot achieve good results. ...

AI Technical Summary

Problems solved by technology

However, the cross-entropy loss function only cares about the accuracy of the prediction probability of positive labels, which leads to low prediction accuracy of the trained model in actual use and cannot achieve good results.

Images