Verification by metamorphic testing of applications that utilize artificial intelligence

• US 10,438,118 B2
• Filed: 09/28/2018
• Issued: 10/08/2019
• Est. Priority Date: 10/09/2017
• Status: Active Grant

• Current Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Original Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Inventors: Dwarakanath, Anurag, Podder, Sanjay, Dubash, Neville, Durg, Kishore P, Ahuja, Manish, Rao, Raghotham M, Sikand, Samarth, Bose Rantham Prabhakara, Jagadeesh Chandra
• Primary Examiner(s): Chen, Alan
• Application Number:
  US16/146,069
• Publication Number:
  US 20190108443A1
• Time in Patent Office: 375 Days
• Field of Search: None
• CPC Class Codes: G06F 11/3684 for test design, e.g. generating new test cases
• CPC Class Codes: G06F 11/3692 for test results analysis
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A device may receive, from a user device, a request to verify a machine learning (ML) application using a metamorphic testing procedure. The device may determine a type of ML process used by the ML application, and may select one or more metamorphic relations (MRs), to be used for performing the metamorphic testing procedure, based on the type of ML process. The device may receive test data to be used to test the ML application, wherein the test data is based on the one or more MRs, and may perform, by using the one or more MRs and the test data, the metamorphic testing procedure to verify one or more aspects of the ML application. The device may generate a report that indicates whether the one or more aspects of the ML application have been verified and may provide the report for display on an interface of the user device.

Claims

1. A method, comprising:

receiving, by a device and from a user device, a request to verify a machine learning (ML) application,

wherein verification of the ML application is to be performed using a metamorphic testing procedure, and

wherein the request includes an indication to validate the metamorphic testing procedure;

determining, by the device, a type of ML process used by the ML application after receiving the request;

selecting, by the device, one or more metamorphic relations (MRs), that are to be used for performing the metamorphic testing procedure, based on the type of ML process used by the ML application;

receiving, by the device, test data that is to be used to test the ML application,

wherein the test data is based on the one or more MRs;

performing, by the device and by using the one or more MRs and the test data, the metamorphic testing procedure to verify one or more aspects of the ML application;

validating, by the device, the metamorphic testing procedure by using a mutation testing technique to determine whether the one or more MRs identify particular errors that were inserted into the ML application for testing;

generating, by the device, a report that includes verification results data indicating whether the one or more aspects of the ML application have been verified and validation results data identifying a result of validating the metamorphic testing procedure; and

providing, by the device, the report for display on an interface of the user device.

2. The method of claim 1, wherein selecting the one or more MRs comprises:

selecting the one or more MRs, from a plurality of MRs, based on the type of ML process used by the ML application and at least one of;

a first indication of whether source code of the ML application has been made accessible to the device, or

a second indication of whether training data used to train the ML application has been made accessible to the device.

3. The method of claim 1, further comprising:

determining, before selecting the one or more MRs, one or more types of data used by the ML application;

selecting one or more tools to be used within an environment that will support performance of the metamorphic testing procedure,

wherein the one or more tools are selected based on the one or more types of data used by the ML application;

configuring the environment to support the one or more tools; and

wherein verifying the one or more aspects of the ML application comprises;

using the one or more MRs and the test data to perform the metamorphic testing procedure in the environment.

4. The method of claim 1, wherein the type of ML process used by the ML application is:

a first type of ML process that uses classification,

a second type of ML process that uses clustering, or

a third type of ML process that uses regression.

5. The method of claim 1, wherein the one or more MRs include at least one of:

a first MR for testing permutations of a class label of a class associated with the ML application,

a second MR for testing permutations of one or more features used to train the ML application,

a third MR for testing permutations that modify the one or more features by a constant value,

a fourth MR for testing permutations of an order in which the one or more features are used to train the ML application,

a fifth MR for testing scaling of the one or more features, or

a sixth MR for testing normalization of the one or more features.

6. The method of claim 1, wherein the ML application includes at least one aspect that is unable to be verified;

and wherein generating the report comprises;

generating one or more recommendations for correcting errors identified within the ML application or for improving accuracy of a data model used for the ML application; and

generating the report that includes the verification results data,

wherein the verification results data includes at least one of;

first data identifying the at least one aspect of the ML application that is unable to be verified,

second data identifying a particular MR, of the one or more MRs, that failed the metamorphic testing procedure, or

third data identifying the one or more recommendations.

7. The method of claim 1, wherein validating the metamorphic testing procedure comprises:

validating the metamorphic testing procedure based on completion of the metamorphic testing procedure;

or

validating the metamorphic testing procedure based on a threshold confidence level of the result of the metamorphic testing procedure not being satisfied.

8. A device, comprising:

one or more memories; and

one or more processors, operatively connected to the one or more memories, to;

receive, from a user device, a request to verify a machine learning (ML) application,

wherein verification of the ML application is to be performed using a metamorphic testing procedure;

determine a type of ML process used by the ML application after receiving the request;

determine one or more types of data used by the ML application after receiving the request;

select one or more metamorphic relations (MRs), that are to be used for performing the metamorphic testing procedure, based on the type of ML process used by the ML application;

generate test data that is to be used to test the one or more MRs;

configure an environment that will support performance of the metamorphic testing procedure after selecting the one or more MRs;

verify one or more aspects of the ML application by using the one or more MRs and the test data to perform the metamorphic testing procedure in the environment;

validate the metamorphic testing procedure by using a mutation testing technique to determine whether the one or more MRs identify particular errors that were inserted into the ML application for testing;

generate a report that includes verification results data indicating whether the one or more aspects of the ML application have been verified and the validation results identifying a result of validating the metamorphic testing procedure; and

provide the report to the user device.

9. The device of claim 8, wherein the one or more processors, when selecting the one or more MRs, are to:

select the one or more MRs, from a plurality of MRs, based on;

the type of ML process used by the ML application,

a first indication of whether source code of the ML application has been made accessible to the device, and

a second indication of whether training data used to train the ML application has been made accessible to the device.

10. The device of claim 8, wherein the one or more types of data used by the ML application include at least one of:

image data,

text data,

audio data, or

multimedia data.

11. The device of claim 8, wherein the one or more processors, when configuring the environment, are to:

select one or more tools to be used within the environment,

wherein the one or more tools are selected based on at least one of;

the type of ML process used by the ML application, or

the one or more types of data used by the ML application; and

configure the environment to support the one or more tools.

12. The device of claim 8, wherein the type of ML process used by the ML application is:

a first type of ML process that uses supervised ML,

a second type of ML process that uses unsupervised ML,

a third type of ML process that uses reinforcement ML.

13. The device of claim 8, wherein the one or more processors, when providing the report to the user device, are to:

provide the report for display on a user interface of the user device,

wherein the report includes the verification results data indicating whether the one or more aspects of the ML application have been verified and at least one of;

data identifying a first group of recommendations associated with correcting errors identified within the ML application, or

data identifying a second group of recommendations associated with improving accuracy of a data model used by the ML application.

14. The device of claim 8, wherein the one or more processors, when validating the metamorphic testing procedure, are to:

validate the metamorphic testing procedure based on completion of the metamorphic testing procedure;

or

validate the metamorphic testing procedure based on a threshold confidence level of the result of the metamorphic testing procedure not being satisfied.

15. A non-transitory computer-readable medium storing instructions, the instructions comprising:

one or more instructions that, when executed by one or more processors of a device, cause the one or more processors to;

receive, from a user device, a request to verify a machine learning (ML) application,

wherein verification of the ML application is to be performed using a testing procedure;

determine a type of ML process used by the ML application after receiving the request;

select one or more properties, that are to be used for performing the testing procedure, based on at least one of;

the type of ML process used by the ML application,

a first indication of whether source code of the ML application has been made accessible to the device, or

a second indication of whether training data used to train the ML application has been made accessible to the device;

generate test data that is to be used to test the one or more properties;

perform, by using the one or more properties and the test data, the testing procedure to selectively verify one or more aspects of the ML application;

validate the testing procedure by using a mutation testing technique to determine whether the one or more properties identify particular errors that were inserted into the ML application for testing;

generate a report that includes verification results data indicating whether the one or more aspects of the ML application have been verified and validation results data identifying a result of validating the testing procedure; and

provide the report for display on an interface of the user device.

16. The non-transitory computer-readable medium of claim 15, wherein the testing procedure is a metamorphic testing procedure and the one or more properties are one or more metamorphic relations (MRs).

17. The non-transitory computer-readable medium of claim 16, wherein the one or more MRs include at least one of:

a first MR for testing permutations of a class label of a class associated with the ML application,

a second MR for testing permutations of one or more features used to train the ML application,

a third MR for testing permutations that modify the one or more features by a constant value,

a fourth MR for testing permutations of an order in which the one or more features are used to train the ML application,

a fifth MR for testing scaling of the one or more features, or

a sixth MR for testing normalization of the one or more features.

18. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to:

determine one or more types of data used by the ML application;

select one or more tools to be used as part of a virtual environment that will support performance of the testing procedure,

wherein the one or more tools are selected based on the one or more types of data;

configure the virtual environment to support the one or more tools; and

wherein the one or more instructions, that cause the one or more processors to selectively verify the one or more aspects of the ML application, cause the one or more processors to;

use the one or more properties and the test data to perform the testing procedure in the virtual environment.

19. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, that cause the one or more processors to generate the report, cause the one or more processors to:

generate one or more recommendations identifying actions to perform,

wherein the actions are selected based on verification of the one or more aspects of the ML, and

wherein the actions include at least one of;

a first action to proceed with deployment of the ML application,

a second action associated with correcting errors identified within the ML application, or

a third action associated with improving accuracy of a data model used for the ML application; and

include, in the report, the one or more recommendations.

20. The non-transitory computer-readable medium of claim 15, wherein the one or more instructions, that cause the one or more processors to validate the testing procedure, cause the one or more processors to:

validate the testing procedure based on completion of the testing procedure;

or

validate the testing procedure based on a threshold confidence level of the result of the testing procedure not being satisfied.