Predicting system status with trustworthy artificial intelligence

Patent metadata

Field	Value
Publication number	US-2024428126-A1
Application number	US-202318339407-A
Country	US
Kind code	A1
Filing date	Jun 22, 2023
Priority date	Jun 22, 2023
Publication date	Dec 26, 2024
Grant date	—

Abstract

One or more systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to training an AI model to predict status of a DBMS. The computer-implemented system can comprise a memory that can store computer executable components. The computer-implemented system can further comprise a processor that can execute the computer executable components stored in the memory, wherein the computer executable components can comprise a data ingestion component that can use testing data of an AI model to generate ingested data by randomly changing one or more records of at least one feature comprised in the testing data, wherein the ingested data can be used to compute a first ratio indicative of inequity of the at least one feature. The computer executable components can further comprise a training component that can train the AI model using at least the first ratio to predict a status of system.

First claim

What is claimed is:

1 . A system, comprising:
a memory that stores computer executable components; and a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise: a data ingestion component that uses testing data of an artificial intelligence (AI) model to generate ingested data by randomly changing one or more records of at least one feature comprised in the testing data based on a specific rule, wherein the ingested data is used to compute a first ratio indicative of inequity of the at least one feature; a runtime deviation check component that compares runtime data of the AI model with the ingested data to identify one or more matching records and generate a list comprising at least the one or more matching records; and a training component that trains the AI model based on an influence factor of the at least one feature, the first ratio of the at least one feature and the list to predict a status of a database management system.

2 . The system of claim 1 , further comprising:
an extraction component that extracts one or more influence factors for one or more respective features of the testing data, wherein the at least one feature is selected for generating the ingested data based on the influence factor being greater than a first threshold.

3 . The system of claim 1 , wherein generating the ingested data comprises generating one or more ingested records for the at least one feature based on one or more existing records for the at least one feature in the testing data.

4 . The system of claim 1 , wherein the first ratio is generated based on an amount of records comprised in the ingested data and an inequity count of the at least one feature determined using the AI model.

5 . The system of claim 4 , wherein the inequity count is determined by comparing prediction results generated by the AI model based on processing one or more records in testing data of the AI model and based on processing one or more ingested records.

6 . The system of claim 1 , wherein the list further comprises prediction results generated by the AI model based on processing the one or more matching records, and wherein the runtime deviation check component uses the prediction results to determine runtime deviation of the AI model.

7 . The system of claim 1 , further comprising:
a trust monitor component that monitors the influence factor for the at least one feature, the first ratio of the at least one feature and the list to generate a second ratio.

8 . The system of claim 7 , further comprising:
a trust model manager component that generates feedback for training the AI model based on at least the first ratio being greater than a threshold or the second ratio being lower than a second threshold.

9 . The system of claim 1 , wherein the AI model is trained to predict the status of the database management system while maintaining inequity and runtime deviation of the AI model below respective thresholds.

10 . A computer-implemented method, comprising:
generating, by a system operatively coupled to a processor, ingested data by randomly changing one or more records of at least one feature comprised in testing data of an AI model based on a specific rule, wherein the ingested data is used to compute a first ratio indicative of inequity of the at least one feature; comparing, by the system, runtime data of the AI model with the ingested data to identify one or more matching records and generate a list comprising at least the one or more matching records; and training, by the system, the AI model based on an influence factor of the at least one feature, the first ratio of the at least one feature and the list to predict a status of a database management system.

11 . The computer-implemented method of claim 10 , further comprising:
extracting, by the system, one or more influence factors for one or more respective features of the testing data, wherein the at least one feature is selected for generating the ingested data based on the influence factor being greater than a first threshold.

12 . The computer-implemented method of claim 10 , further comprising:
generating, by the system, generating one or more ingested records for the at least one feature based on one or more existing records for the at least one feature in the testing data.

13 . The computer-implemented method of claim 10 , further comprising:
generating, by the system, the first ratio based on an amount of records comprised in the ingested data and an inequity count of the at least one feature determined using the AI model.

14 . The computer-implemented method of claim 13 , further comprising:
determining, by the system, the inequity count by comparing prediction results generated by the AI model based on processing one or more records in testing data of the AI model and based on processing one or more ingested records.

15 . The computer-implemented method of claim 10 , wherein the list further comprises prediction results generated by the AI model based on processing the one or more matching records, and wherein the prediction results are used to determine runtime deviation of the AI model.

16 . The computer-implemented method of claim 9 , further comprising:
monitoring, by the system, the influence factor for the at least one feature, the first ratio of the at least one feature and the list to generate a second ratio; and generating, by the system, feedback for training the AI model based on at least the first ratio being greater than a threshold or the second ratio being lower than a second threshold.

17 . A computer program product for training an AI model to improve robustness of the AI model to training inequity and runtime deviation, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
generate, by the processor, ingested data by randomly changing one or more records of at least one feature comprised in testing data of an AI model based on a specific rule, wherein the ingested data is used to compute a first ratio indicative of inequity of the at least one feature; compare, by the processor, runtime data of the AI model with the ingested data to identify one or more matching records and generate a list comprising at least the one or more matching records; and train, by the processor, the AI model based on an influence factor of the at least one feature, the first ratio of the at least one feature and the list to predict a status of a database management system.

18 . The computer program product of claim 17 , wherein the program instructions are further executable by the processor to cause the processor to:
extract, by the processor, one or more influence factors for one or more respective features of the testing data, wherein the at least one feature is selected for generating the ingested data based on the influence factor being greater than a first threshold.

19 . The computer program product of claim 17 , wherein the program instructions are further executable by the processor to cause the processor to:
generate, by the processor, one or more ingested records for the at least one feature according to one or more existing records for the at least one feature in the testing data.

20 . The computer program product of claim 17 , wherein the program instructions are further executable by the

Assignees

• IBM

Inventors

• Shang Guan Ming Qiao
• Zeng Mai
• Wan Meng
• Dong Xin Xin
• Sun sheng yan
• Song Wei
• Liu wen zhong

Classifications

• G06N20/00 Primary

  Machine learning · CPC title