Sample weight allocation method, model training method, electronic equipment and storage medium

Abstract

The invention provides a sample weight allocation method. The method comprises the following steps: acquiring training samples, wherein the training samples include a positive sample set and a negative sample set; calculating the distance of each positive sample couple in the positive sample set, and the distance of each negative sample couple in the negative sample set; determining distance distribution of the positive sample set according to the distance of each positive sample couple in the positive sample set; determining distance distribution of the negative sample set according to the distance of each positive sample couple in the negative sample set; and determining weight distribution of the training samples based on the distance distribution of the positive sample set and the distance distribution of the negative sample set. The invention further provides a model training method, electronic equipment and a storage medium. According to the sample weight allocation method disclosed by the invention, the weight of the sample couples with wrong classification can be increased, and contribution of the samples with classification errors to targeted loss is increased in the modeltraining process, so that model parameters can be well corrected, and the expression ability of the model parameters is improved.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a sample weight distribution method, a model training method, electronic equipment and a storage medium. Background technique [0002] In the field of machine learning, in the training of models (such as feature extraction models, face feature expression models, etc.), loss functions are divided into two categories. The first category is based on classification measurements. Since the features are not directly measured, the performance is limited; The other type is an end-to-end method directly oriented to feature measurement. This type of method can converge well because it needs to select a sample network with an appropriate degree of difficulty. Existing methods mainly use the following two methods to obtain samples with appropriate difficulty: first, after the model is trained to a certain stage, according to the characteristic expression of the model, some samples with...

AI Technical Summary

Problems solved by technology

Existing methods mainly use the following two methods to obtain samples with appropriate difficulty: first, after the model is trained to a certain stage, according to the characteristic expression of the model, some samples with moderate difficulty are selected, which is cumbersome to operate, and With the training of the model, the difficulty of the selected samples changes, and the original offline selected samples are no longer representative, and cannot fully express the characteristics of the subsequent added samples.

Second, in the process of model training, samples with moderate difficulty are selected according to the model for each training. Although the training samples selected by this method are representative and can effectively improve the expressive ability of the model, the computing resources required are too large. Large, difficult to achieve in actual model training

Images