Method and system for evaluating and optimizing a neural network model

Abstract

The invention provides a method and system for evaluating and optimizing a neural network model. The method comprises: selecting training-valuable data based on an uncertainty index through active learning, wherein the uncertainty index is used to reflect the classification ability of the trained neural network model for the data; constructing a test set based on the selected data and Using the test set to obtain the effectiveness evaluation result of the neural network model; according to the effectiveness evaluation result, optimize the training sample set of the neural network model and update the neural network model until the desired neural network model is obtained . The method and system of the present invention can adapt the neural network model to individual requirements and be easier to supervise.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a method and system for evaluating and optimizing a neural network model. Background technique [0002] In recent years, artificial intelligence (AI) has developed rapidly due to its excellent algorithm performance. However, most of these algorithms are still in the scientific research stage, and there are still certain limitations in practical applications. For example, in the medical field, there is currently no general neural network model that can be applied to all hospitals. The reason is that government supervision is difficult, and the data of each hospital is specific. The rapid development of AI relies on the efficient performance of convolutional neural networks, which are much better than many traditional algorithms and can be applied to many fields. However, the convolutional neural network requires a large amount of labeled data and has certain specificity, th...

AI Technical Summary

Problems solved by technology

Doing so discourages government scrutiny and adapts the model to new data and different hospitals

In the prior art, the evaluation methods for the effectiveness of the neural network model cannot meet the individual requirements of different hospitals. These methods cannot make the original model suitable for different hospitals, nor can the original model adapt to new data. At the same time, it is not easy for the government to supervise the effectiveness of the neural network model

Images