Machine learning sample weight adjustment method and device and storage medium

Abstract

The invention relates to a machine learning sample weight adjustment method and device and a storage medium. The machine learning sample weight adjustment method comprises the steps of determining anerror between a sample prediction value and a sample label value of each sample after each round of model training in a multi-round iterative training process of a machine learning model for a scene in which the number of different types of samples is unbalanced; and dynamically adjusting the current sample weight of each sample after model training according to the error between the sample prediction value and the sample label value, and taking the current sample weight as the sample weight of the next round of model training. According to the method, dynamic and fine adjustment of the weightof the learning sample can be realized through a heuristic iterative thought.

Description

technical field [0001] The present disclosure relates to its field of learning, in particular to a method and device for adjusting machine learning sample weights, and a storage medium. Background technique [0002] With the continuous growth of computer computing power, data volume, and data dimensions, machine learning has penetrated into all aspects of modern life and has become an important support for various Internet services: in search, recommendation, navigation, anti-fraud, etc. It is due to the continuous development and iteration of various machine learning algorithms that users can get better and better experience. According to whether the training data is marked, machine learning methods can be simply divided into supervised learning (Supervised Machine Learning) and unsupervised learning (Unsupervised Machine Learning). The current mainstream machine learning method is supervised learning. For supervised learning methods, the sample label (Label) is extremely ...

AI Technical Summary

Problems solved by technology

[0004] The inventor found through research that: for the problem of sample imbalance, related technologies solve the problem from the data level and the algorithm level: first, adjust the sample ratio from the data volume level by upsampling or downsampling, where upsampling usually refers to synthesis or copying Technology generates more minority class samples, and generates more minority class samples through synthesis technology. Downsampling usually refers to reducing the majority class samples through sampling technology to achieve the purpose of balancing the number of samples of each type; second, by increasing the number of minority class samples The sample weight (or reduce the learning weight of most samples) makes the model bias towards few samples in gradient solution or loss calculation

Images