Public policy rule enhancement of machine learning/artificial intelligence solutions

Patent metadata

Field	Value
Publication number	US-11803765-B2
Application number	US-201916427753-A
Country	US
Kind code	B2
Filing date	May 31, 2019
Priority date	May 31, 2019
Publication date	Oct 31, 2023
Grant date	Oct 31, 2023

Abstract

A method includes creating one or more first policy shims to be applied to a ML/AI module, applying the one or more first policy shims to an input or an output of the ML/AI module and executing the ML/AI module on a data set in response to the applying step. The one or more first policy shims includes an input policy shim and an output policy shim and the applying step includes applying the input policy shim to the data set prior to the executing step and applying the output policy shim to an output of the executing step.

First claim

The invention claimed is:

1. A method comprising:
creating, by a processor, one or more first policy shims and one or more substitute policy shims for an input or an output of a machine learning module;
determining whether the one or more first policy shims comprises an input policy shim, wherein the input policy shim filters data supplied to the input of the machine learning module;
applying, by the processor, the one or more first policy shims to an input data set provided to the input of the machine learning module to filter the input data set responsive to a determination that the one or more first policy shims comprises the input policy shim;
applying, by the processor, the one or more substitute policy shims to the input data set provided to the input of the machine learning module responsive to a second determination of a first external input that the one or more substitute policy shims should override the input policy shim;
executing, by the processor, the machine learning module on a data set produced in response to the applying steps, thereby producing an output data set; and
applying, by the processor, the one or more first policy shims to the output data set produced by the machine learning module responsive to a third determination that the one or more first policy shims comprises an output policy shim.

2. The method of claim 1 wherein the machine learning module is one of a plurality of nested modules and the creating step includes creating one or more additional policy shims for at least one of the plurality of nested modules and wherein the applying steps apply the one or more additional policy shims to the one of the plurality of nested modules and wherein the executing step includes executing the plurality of nested modules in response to the applying steps.

3. The method of claim 2 wherein additional one or more data sets are inserted into the plurality of nested modules prior to the executing step.

4. The method of claim 1 , further comprising overriding the output policy shim and applying the one or more substitute policy shims instead of the output policy shim.

5. The method of claim 1 further comprising storing, by the processor, the one or more first policy shims in a database and retrieving the one or more first policy shims prior to the applying steps.

6. The method of claim 1 wherein the one or more first policy shims are a filter configured to act on the data set.

7. The method of claim 1 wherein the one or more first policy shims are a set of business rules.

8. The method of claim 4 , wherein the overriding the output policy shim is responsive to a fourth determination associated with a second external input.

9. The method of claim 8 wherein the first external input is one of a weather alert and an emergency alert.

10. The method of claim 8 wherein the first external input is one or more business rules.

11. The method of claim 1 wherein the machine learning module is modified based on the executing steps and whereby the one or more first policy shims are no longer used.

12. An apparatus comprising:
a policy database configured for storing a plurality of policies;
an input-output interface;
a processor coupled to the input-output interface wherein the processor is coupled to a memory, the memory having stored thereon executable instructions that when executed by the processor cause the processor to effectuate operations comprising:
creating one or more first policy shims from the plurality of policies and one or more substitute policy shims;
determining whether the one or more first policy shims comprises an input policy shim wherein the input policy shim filters data supplied to an input of a machine learning module;
applying the input policy shim to an input data set provided to the input of the machine learning module to filter the input data set responsive to a determination that the one or more first policy shims comprises the input policy shim;
applying the one or more substitute policy shims to the input data set provided to the input of the machine learning module responsive to a second determination of a first external input indicating the one or more substitute policy shims should override the input policy shim; and
executing the machine learning module on a data set produced in response to the applying steps thereby producing an output data set; and
applying the one or more first policy shims to the output data set produced by the machine learning module responsive to a third determination that the one or more first policy shims comprises an output policy shim.

13. The apparatus of claim 12 wherein the operations further comprise creating a first substitute policy shim and receiving, via the input-output interface, an input from an external source and overriding the one or more first policy shims with the first substitute policy shim.

14. The apparatus of claim 13 wherein the first substitute policy shim is a filter or a business rule.

15. The apparatus of claim 12 wherein the operations further comprise creating a safety policy shim and receiving, via the input-output interface, an input from an external source and overriding the one or more first policy shims with the safety policy shim.

16. The apparatus of claim 15 wherein the safety policy shim is a filter or a business rule to be applied based on a safety concern.

17. A non-transitory computer-readable storage medium comprising instructions that, when executed by a processor, facilitate a performance of operations, the operations comprising:
creating one or more first policy shims and one or more substitute policy shims to be applied to a machine learning module;
applying the one or more first policy shims to an input data set provided to an input of the machine learning module to filter the input data set responsive to a determination that the one or more of the first policy shims comprise input policy shims that filter data supplied to the input of the machine learning module;
applying the one or more substitute policy shims to the input data set provided to the input of the machine learning module responsive to a second determination of a first external input that the one or more substitute policy shims should override the one or more first policy shims;
executing the machine learning module on a second data set produced in response to the applying steps, thereby producing an output data set; and
applying the one or more first policy shims to the output data set produced by the machine learning module responsive to a third determination that the one or more first policy shims comprises an output policy shim.

18. The non-transitory computer-readable storage medium of claim 17 , wherein the one or more first policy shims comprise a filter configured to act on the second data set.

Assignees

• At & T Ip I Lp

Inventors

• Aaron Jeffrey
• Fan James

Classifications

• G06N5/027 Primary

  Frames · CPC title

• G06F9/44526

  Plug-ins; Add-ons · CPC title

• G06N20/00

  Machine learning · CPC title

• G06Q10/067 Primary

  Enterprise or organisation modelling · CPC title