Classifier, neural network model training method, data processing device, and medium

Abstract

The invention discloses a classifier, a neural network model training method, data processing equipment and a medium. A classifier comprises: a weighted loss value calculation means for obtaining a weighted loss value of a training sample on each classification category by means of a prediction confidence output from a neural network model; a Parameter adjusting device, wherein the weighted loss value is reversely output to the parameter adjusting device; Therefore, the parameters of the neural network model are adjusted based on the weighted loss value, and the traininig of the neural networkmodel of the classifer can be completed. wherein the weighted loss value calculation device is used for adding a weighting factor into a loss function used for measuring the prediction confidence coefficient of a training sample and the loss cost marked with a true value, so that the weighted loss value of the training sample on each classification category is obtained through calculation. By adding a weighting factor during loss value calculation, parameters of the neural network model can be adjusted, the network model can be optimized, and the recognition precision of the classifier can beimproved.

Description

technical field [0001] The invention relates to classification technology, in particular to a classifier, a neural network model training method, data processing equipment and media. Background technique [0002] Deep learning is currently the mainstream method with excellent accuracy and widely used in the field of computer pattern recognition. In the application process of the deep learning model, it is first necessary to train the deep learning model on a large training data set, through a custom loss function and a backpropagation algorithm (such as the gradient descent of the backpropagation algorithm for multiple iterations) function) to learn iteratively optimal neural network model parameters, and then deploy it to practical applications for classification or prediction after verifying that the target is reached on the verification set. [0003] In the actual data training of deep learning, the problem of low classification accuracy often occurs due to the unbalance...

AI Technical Summary

Problems solved by technology

[0005] In order to solve one of the above problems, the present invention provides a classifier, a neural network model training method, data processing equipment and media, in order to overcome the problem of low recognition accuracy of the classifier, thereby improving its recognition accuracy

Images