A deep learning model tuning management method, device, equipment and medium

Abstract

The invention discloses a method for tuning and managing a deep learning model, which includes: in response to the number of pictures in a data set that needs to be predicted is greater than a set number threshold, the data set is divided into several sub-data sets on average, and each sub-data set is processed by a deep learning model. The data set is predicted to obtain the first prediction data; the judgment data of each sub-data set is obtained through manual prediction and judgment; the Kalman filter is performed on the first prediction data and judgment data of each sub-data set to obtain the first prediction data of each sub-data set Two prediction data; obtain the error rate of the deep learning model according to the first prediction data and the second prediction data; trigger the retraining of the deep learning model in response to the error rate exceeding the error rate threshold. The invention also discloses a device, equipment and medium. The principle of the invention is easy to understand, easy to operate and realize, and the accuracy is guaranteed, and the function of automatic model retraining and tuning is optimized.

Description

technical field [0001] The present invention relates to the field of image processing, and more particularly, to a deep learning model tuning management method, apparatus, device and medium. Background technique [0002] At present, the retraining management of the prediction model can be divided into the following two types: one is to compare the prediction results of the model with the manual annotation results, and accumulate the prediction error results identified by humans, that is, the number of pictures. When the number of errors reaches a threshold, all the positive and negative sample sets accumulated in the previous prediction will be used as the training set to retrain the model (for example, if the threshold is 200 wrongly predicted pictures, the first 1000 pictures are predicted incorrectly by 100 pictures, and the second The second 2000 pictures are predicted wrong by 150 pictures, then after the second prediction is over, the previous 3000 pictures will be use...

AI Technical Summary

Problems solved by technology

[0003] For the management of the retraining model, the retraining period and triggering situation are more important. Too many retraining times indicate that the model is difficult to meet the accuracy requirements of the manager, and retraining consumes time and computing power. If the number of retraining is too small, the model will not be updated for a long time, and the prediction accuracy for new situations that the model has not learned will be low, and the total number of prediction sets for each prediction is too large or too small, which will affect trigger for retraining

Images