Machine learning model training method and device and readable storage medium

Abstract

The invention discloses a training method and device of a machine learning model and a readable storage medium. The method comprises the steps of obtaining a plurality of first training samples; performing first random enhancement processing on the first training sample to obtain a second training sample; after it is judged that the second training sample meets a preset enhancement condition based on a preset total enhancement number, second random enhancement processing is carried out on the second training sample to obtain a third training sample, and the preset total enhancement number is a divisor of the batch processing number; and training the machine learning model by using the training samples to obtain a trained machine learning model, the training samples including a second training sample or a third training sample, and the batch processing number being the number of the training samples input to the machine learning model each time. In this way, the convergence speed of the model can be increased.

Description

technical field [0001] The present application relates to the technical field of machine learning, in particular to a training method, device and readable storage medium of a machine learning model. Background technique [0002] In order to obtain a better model training effect, some programs perform data cleaning operations before model training, that is, filter the acquired data to obtain selected data to form training data for model training, thereby improving the efficiency of model training , but it may make the model easy to overfit, resulting in insufficient ability of model detection or classification in real scenes; some schemes randomly select part of the original training data (such as: image, text or voice) at each iteration, and then perform random Enhancement methods such as crop, resize or random tone adjustment are used to improve the generalization ability of the model, but in continuous iterative training, the training data may still be biased towards a cer...

AI Technical Summary

Problems solved by technology

[0002] In order to obtain a better model training effect, some programs perform data cleaning operations before model training, that is, filter the acquired data to obtain selected data to form training data for model training, thereby improving the efficiency of model training , but it may make the model easy to overfit, resulting in insufficient ability of model detection or classification in real scenes; some schemes randomly select part of the original training data (such as: image, text or voice) at each iteration, and then perform random Enhancement methods such as crop, resize or random tone adjustment are used to improve the generalization ability of the model, but in continuous iterative training, the training data may still be biased towards a certain feature, resulting in the weight of the model being in the In a certain period of time, all are updated for a certain feature, so that the loss fluctuates greatly during the entire training process, which not only leads to an infinite increase in training time, but also may cause the training to fail to converge.

Images