Deployment of Machine Learning Models for Discernment of Threats

• US 20170357807A1
• Filed: 06/06/2017
• Published: 12/14/2017
• Est. Priority Date: 06/08/2016
• Status: Active Grant

• Current Assignee: Cylance Inc. (Blackberry Limited)
• Original Assignee: Cylance Inc. (Blackberry Limited)
• Inventors: Harms, Kristopher William, Song, Renee, Rajamani, Raj, Rusell, Braden, Sohn, Alice, Ipsen, Kiefer
• Granted Patent: US 10,372,913 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 21/51 at application loading time, e.g. accepting, rejecting, starting or inhibiting executable software based on integrity or source reliability
• CPC Class Codes: G06F 21/562 Static detection
• CPC Class Codes: G06F 21/566 Dynamic detection, i.e. detection performed at run-time, e.g. emulation, suspicious activities
• CPC Class Codes: G06F 21/568 eliminating virus, restoring damaged files
• CPC Class Codes: G06F 2221/033 Test or assess software
• CPC Class Codes: G06F 3/048 Interaction techniques based on graphical user interfaces [GUI]
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: H04L 63/1425 Traffic logging, e.g. anomaly detection
• CPC Class Codes: H04L 63/1441 Countermeasures against malicious traffic countermeasures against attacks on cryptographic mechanisms H04L9/002

Abstract

A mismatch between model-based classifications produced by a first version of a machine learning threat discernment model and a second version of a machine learning threat discernment model for a file is detected. The mismatch is analyzed to determine appropriate handling for the file, and taking an action based on the analyzing. The analyzing includes comparing a human-generated classification status for a file, a first model version status that reflects classification by the first version of the machine learning threat discernment model, and a second model version status that reflects classification by the second version of the machine learning threat discernment model. The analyzing can also include allowing the human-generated classification status to dominate when it is available.

Claims

1. A method for implementation by one or more data processors forming part of at least one computing device, the method comprising:

detecting, by at least one data processor, a mismatch between model-based classifications produced by a first version of a computer-implemented machine learning threat discernment model and a second version of a computer-implemented machine learning threat discernment model for a file;

analyzing, by at least one data processor, the mismatch to determine appropriate handling for the file, the analyzing comprising comparing a human-generated classification for a file, a first model version status that reflects classification by the first version of the machine learning threat discernment model, and a second model version status that reflects classification by the second version of the machine learning threat discernment model, the analyzing further comprising allowing the human-generated classification to dominate when it is available; and

taking, by at least one data processor, an action based on the analyzing.

2. The computer-implemented method of claim 1, wherein the action comprises:

presenting, by at least one data processor, the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file;

quarantining, by at least one data processor, a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file; and

presenting, by at least one data processor, the file via the user interface functionality of the administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

3. The computer-implemented method of claim 1, wherein the action comprises:

presenting, by at least one data processor, the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file.

4. The computer-implemented method of claim 1, wherein the action comprises:

quarantining, by at least one data processor, a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file.

5. The computer-implemented method of claim 1, wherein the action comprises:

presenting, by at least one data processor, the file via user interface functionality of an administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

6. The computer-implemented method of claim 1, wherein each of the first version of the machine learning threat discernment model and the second version of the machine learning threat discernment model for a file output a respective threat score, wherein the mismatch is based on differences between the threat scores.

7. The computer-implemented method of claim 1, wherein the action comprises:

executing or otherwise accessing code that generated the respective threat scores without quarantining.

8. The computer-implemented method of claim 1, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on different and remote computing systems.

9. The computer-implemented method of claim 1, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on the same computing system.

10. A computer program product comprising a non-transitory machine-readable medium storing instructions that, when executed by at least one data processor forming part of at least one computing device, cause the at least one data processor to perform operations comprising:

detecting a mismatch between model-based classifications produced by a first version of a computer-implemented machine learning threat discernment model and a second version of a computer-implemented machine learning threat discernment model for a file;

analyzing the mismatch to determine appropriate handling for the file, the analyzing comprising comparing a human-generated classification for a file, a first model version status that reflects classification by the first version of the machine learning threat discernment model, and a second model version status that reflects classification by the second version of the machine learning threat discernment model, the analyzing further comprising allowing the human-generated classification to dominate when it is available; and

taking an action based on the analyzing.

11. The computer program product of claim 10, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file;

quarantining a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file; and

presenting the file via the user interface functionality of the administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

12. The computer program product of claim 10, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file.

13. The computer program product of claim 10, wherein the action comprises:

quarantining a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file.

14. The computer program product of claim 10, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

15. The computer program product of claim 10, wherein each of the first version of the machine learning threat discernment model and the second version of the machine learning threat discernment model for a file output a respective threat score, wherein the mismatch is based on differences between the threat scores.

16. The computer program product of claim 10, wherein the action comprises:

executing or otherwise accessing code that generated the respective threat scores without quarantining if a difference in the threat scores is below a pre-defined threshold.

17. The computer program product of claim 10, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on different and remote computing systems.

18. The computer program product of claim 10, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on the same computing system.

19. A system comprising:

at least one data processor; and

memory storing instructions which, when executed by the at least one data processor, result in operations comprising;

detecting a mismatch between model-based classifications produced by a first version of a computer-implemented machine learning threat discernment model and a second version of a computer-implemented machine learning threat discernment model for a file;

analyzing the mismatch to determine appropriate handling for the file, the analyzing comprising comparing a human-generated classification for a file, a first model version status that reflects classification by the first version of the machine learning threat discernment model, and a second model version status that reflects classification by the second version of the machine learning threat discernment model, the analyzing further comprising allowing the human-generated classification to dominate when it is available; and

taking an action based on the analyzing.

20. The system of claim 19, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file;

quarantining a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file; and

presenting the file via the user interface functionality of the administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

21. The system of claim 19, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a choice of creating a new human-generated classification of the file as safe or allowing the second model version classification to govern when the file is deemed safe by the first model version but unsafe by the second model version, and when the human-generated classification is unavailable for the file.

22. The system of claim 19, wherein the action comprises:

quarantining a file deemed unsafe by the first version of the machine learning threat discernment model but safe by the second version of the machine learning threat discernment model when the human-generated classification is unavailable for the file.

23. The system of claim 19, wherein the action comprises:

presenting the file via user interface functionality of an administrative interface with a second choice of designating the file for local analysis and/or allowing the second model version classification to continue to block use of the file until the human-generated classification becomes available when the file is deemed unsafe by the second model version and the human human-generated classification is unavailable and when the first model version has not previously classified the file.

24. The system of claim 19, wherein each of the first version of the machine learning threat discernment model and the second version of the machine learning threat discernment model for a file output a respective threat score, wherein the mismatch is based on differences between the threat scores.

25. The system of claim 19, wherein the action comprises:

executing or otherwise accessing code that generated the respective threat scores without quarantining.

26. The system of claim 19, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on different and remote computing systems.

27. The system of claim 20, wherein the first version of a computer-implemented machine learning threat discernment model and the second version of a computer-implemented machine learning threat discernment model for the file are on the same computing system.