Calibration method for artificial intelligence model

Abstract

Detection instruments based on artificial intelligence are increasingly popularized, but calibration of the artificial intelligence detection instruments is a difficult problem which is not well solved all the time, and lack of a calibration method of an artificial intelligence model is a key causing the passive situation. After an artificial intelligence model (AI model) is established, an unknown point is predicted by using the AI model, which cannot only meet the value of a prediction point, and more importantly, a confidence interval of the prediction point under a certain confidence can be given. However, in an existing AI model building method, a calculation formula of an AI model confidence interval is generally not given. The invention provides a calibration method for an artificial intelligence model, and the method comprises the steps: constructing an error set of a prediction result through a K-fold cross validation method through a training sample set of an AI model; and when the number of training samples of the AI model is greater than 30, combining the Z test rule with a principal component method independent of the AI model to determine a confidence interval of theAI model under a given confidence; and when the number of the training samples of the AI model does not exceed 30, combining the t test rule with a principal component method independent of the AI model to determine a confidence interval of the AI model under a given confidence. Compared with the prior art, the technical scheme provided by the invention has the beneficial effects that through theAI model, not only can the value of the prediction point be given, but also the confidence interval of the prediction point under a certain confidence coefficient can be given, and a foundation is laid for the calibration of an artificial intelligence detection instrument.

Description

technical field [0001] The invention relates to the field of metrology, and more specifically, to a calibration method for artificial intelligence models. Background technique [0002] Artificial intelligence is playing an increasingly important role in social and economic development. With the deep integration of artificial intelligence and virtual instruments, more intelligent testing instruments are ushering in an opportunity for vigorous development. Calibration is an indispensable core for any instrument However, in the current AI model modeling methods, the formula for calculating the confidence interval of the AI ​​model is generally not given, which leads to people being able to use the AI ​​model to predict unknown points after building an artificial intelligence model (AI model). Only the value of the predicted point can be obtained, but the confidence interval of the predicted point under a certain confidence level cannot be given. This defect seriously restricts ...

AI Technical Summary

Problems solved by technology

[0002] Artificial intelligence is playing an increasingly important role in social and economic development. With the deep integration of artificial intelligence and virtual instruments, more intelligent testing instruments are ushering in an opportunity for vigorous development. Calibration is an indispensable core for any instrument However, in the current AI model modeling methods, the formula for calculating the confidence interval of the AI ​​model is generally not given, which leads to people being able to use the AI ​​model to predict unknown points after building an artificial intelligence model (AI model). Only the value of the prediction point can be obtained, but the confidence interval of the prediction point under a certain confidence level cannot be given. This defect seriously restricts the development space of more intelligent detection instruments

Images