Classification network training method and device, computer equipment and storage medium

Abstract

The invention relates to a classification network training method and a device, computer equipment and a storage medium. The method comprises the steps of determining a first weight factor of each category of sample data in a sample set according to the number of each category of sample data in the sample set; determining a second weight factor of each piece of sample data according to the gradient value of each piece of sample data in the sample set; and determining a loss value of the classification network according to the first weight factor, the second weight factor, the label corresponding to each piece of sample data and the classification result corresponding to each piece of sample data, and updating parameters in the classification network according to the loss value. According to the method, the value of the loss function containing the first weight factor and the second weight factor is adopted to train the classification network; and meanwhile, the problems of class imbalance of the training samples and sample data imbalance caused by difficult samples or noise samples are solved, the robustness of the trained classification network is improved, and the training efficiency of training the classification network is improved.

Description

technical field [0001] The present application relates to the technical field of deep learning, and in particular to a training method, device, computer equipment and storage medium for a classification network. Background technique [0002] With the rapid development and maturity of deep learning methods, more and more scenarios need to use deep learning methods to solve related technical problems, and deep learning methods are based on massive amounts of data. The phenomenon of data imbalance is becoming more and more obvious. [0003] In recent years, a large number of methods to solve data imbalance have emerged, for example, downsampling data with many categories to reduce the amount of data in this category. Or, perform upsampling processing on data with few categories, or perform augmentation processing on data of this category, so as to increase the amount of data of this category. [0004] However, the above-mentioned methods for solving data imbalance usually hav...

AI Technical Summary

Problems solved by technology

[0004] However, the above-mentioned methods to solve data imbalance usually have the problem that the model is unstable during the training process, which leads to the problem that the training cannot be completed normally.

Images