Neural network training method and device, and electronic equipment

Abstract

The invention discloses a neural network training method and device, and electronic equipment. The method comprises the following steps: obtaining a feature map of a labeled data set from the input labeled data set by a neural network; inputting the feature map of the data set with the label into a Softmax activation function of the neural network to obtain a probability output value of a positioncorresponding to a label; calculating a weighted value based on a monotonically increasing convex function of the probability output value, and weighting an original loss function value of the neuralnetwork by using the weighted value to obtain a new loss function value; and updating parameters of the neural network based on the new loss function value. In this way, the robustness of the neuralnetwork to the label noise can be improved.

Description

technical field [0001] The present application relates to the field of deep learning, and more specifically, to a neural network training method, a neural network training device, and electronic equipment. Background technique [0002] At present, in the field of deep learning, the common work is to use labeled data to train neural networks for classification, regression or other purposes. This method of training model learning rules is generally called supervised learning. [0003] In supervised learning, the quality of the labels corresponding to the training data is crucial to the learning effect. If the wrong labeled data is used for learning, it is impossible to train an effective model. At the same time, since the neural network used in deep learning is often complex in structure, in order to obtain a good learning effect, there are also high requirements for the amount of labeled training data. [0004] However, in the process of mass data labeling, label noise will...

AI Technical Summary

Problems solved by technology

If the labeled data used for learning is wrong, it is impossible to train an effective model

At the same time, due to the complex structure of the neural network used in deep learning, in order to obtain a good learning effect, there are also high requirements for the number of labeled training data.

[0004] However, in the process of mass data labeling, label noise will inevitably be generated, which will seriously reduce the performance of the model during the training process.

This is because the work of labeling data needs to be done manually in many scenarios. Massive, high-quality labels are time-consuming and labor-intensive, and relatively expensive economically.

Images