System method and code for un-learning an individual from an artificial intelligence (AI) model

1 . An Information Handling System (IHS), comprising:

at least one processor; and

at least one memory coupled to the at least one processor, the at least one memory configured with program instructions stored thereon that, upon execution by the at least one processor, cause the IHS to:

receive one or more individual-specific raw data instances associated with an individual, wherein the individual comprises a multi-user account;

receive a plurality of other raw data instances associated with individuals other than the individual;

receive a plurality of individual-specific resource data instances comprising measurements captured by sensors configured in the IHS;

perform a native artificial intelligence (AI) process to generate one or more native features in accordance to the received other raw data instances, the individual-specific raw data instances associated with the multi-user account, and the individual-specific resource data instances, and generate a trained native AI model from the one or more native features, the trained native AI model configured to generate one or more profile recommendations determined to optimize IHS performance and energy usage, wherein the one or more profile recommendations are determined by the trained native AI model as a function of IHS performance and energy usage feature input derived from telemetry data that includes measurements captured by the sensors configured in the IHS, wherein the one or more profile recommendations comprise one or more of: adjust one or more of a frame rate of a display resource, a refresh rate of a display resource, a computational frame rate of a GPU resource, an overclocking or underclocking level of a CPU resource, or a write optimized setting or a read optimized setting of a storage resource;

perform a differential AI process to generate one or more differential features in accordance to the received other raw data instances, and a differential model from the one or more differential features, wherein the differential features are generated without the individual-specific raw data instances; and

in response to an event trigger that indicates a request to remove influence of the individual from the trained native AI model, determine whether the influence of the individual on the trained native AI model was a high level of influence defined as at least 3 percent, based at least in part on;

compare the native features against the differential features to determine one or more feature influence levels of the individual-specific raw data instance on the native features;

compare the trained native AI model against the differential model to determine a model influence level of the individual-specific raw data instance on the trained native AI model;

compare at least one of the one or more feature influence levels, or the model influence level, to a specified threshold; and

in response to a determination that the influence of the individual on the trained native AI model was the high level of influence:

replace the native features and trained native AI model with the differential features and the differential model; and

delete the individual-specific raw data instances and the individual-specific resource data instances in accordance to the model influence level.

2 . The IHS of claim 1 , wherein the event trigger further comprises a user request, and wherein the instructions are further executed to determine the one or more feature influence levels, determine the model influence level, and compare at least one of the feature influence levels or the model influence level against the specified threshold in response to the user request to delete the individual-specific raw data instances.

3 . The IHS of claim 2 , wherein the instructions are further executed to:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, delete the native AI model and the individual-specific raw data instances, and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, delete the individual-specific raw data instances without deletion of the native AI model.

4 . The IHS of claim 2 , wherein the instructions are further executed to:

determine whether the differential features associated with the individual are used in the native AI model; and

when the differential features are not used in the native AI model, delete the individual-specific raw data instances without deletion of the native AI model.

5 . The IHS of claim 2 , wherein the IHS comprises a multi-user IHS that administers the multi-user account, and the instructions are further executed to:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, delete the native AI model and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, leave the native AI model intact.

6 . The IHS of claim 5 , wherein the native AI process further comprises a plurality of AI services, and the instructions are further executed to:

for each of the AI services, compare the feature influence levels or the model influence level against the specified threshold when the AI service has used the individual-specific raw data instances of the individual.

7 . The IHS of claim 1 , wherein the differential AI process is provided as a cloud-based service.

8 . A method comprising:

receiving, using instructions stored in at least one memory and executed by at least one processor, one or more individual-specific raw data instances associated with an individual, wherein the individual comprises a multi-user account;

receiving, using the instructions, a plurality of other raw data instances associated with individuals other than the individual;

receiving, using the instructions, a plurality of individual-specific resource data instances comprising measurements captured by sensors configured in an Information Handling System (IHS);

performing, using the instructions, a native artificial intelligence (AI) process to generate one or more native features according to the received other raw data instances, the individual-specific raw data instances associated with the multi-user account, and the individual-specific resource data instances, and generate a trained native AI model from the one or more native features, the trained native AI model configured to generate one or more profile recommendations determined to optimize IHS performance and energy usage, wherein the one or more profile recommendations are determined by the trained native AI model as a function of IHS performance and energy usage feature input derived from telemetry data including measurements captured by the sensors configured in the IHS, wherein the one or more profile recommendations comprise one or more of: adjusting one or more of a frame rate of a display resource, a refresh rate of a display resource, a computational frame rate of a GPU resource, an overclocking or underclocking level of a CPU resource, or a write optimized setting or a read optimized setting of a storage resource;

performing, using the instructions, a differential AI process to generate one or more differential features according to the received other raw data instances, and generate a differential model from the one or more differential features, wherein the differential features are generated without the individual-specific raw data instances; and

in response to an event trigger indicating a request to remove influence of the individual from the trained native AI model, determining, using the instructions, whether the influence of the individual on the trained native AI model was a high level of influence defined as at least 3 percent, based at least in part on:

comparing, using the instructions, the native features against the differential features to determine one or more feature influence levels of the individual-specific raw data instance on the native features;

comparing, using the instructions, the trained native AI model against the differential model to determine a model influence level of the individual-specific raw data instance on the trained native AI model;

comparing at least one of the one or more feature influence levels, or the model influence level, to a specified threshold; and

in response to determining that the influence of the individual on the trained native AI model was the high level of influence:

replacing, using the instructions, the native features and trained native AI model with the differential features and the differential model; and

deleting, using the instructions, the individual-specific raw data instances and the individual-specific resource data instances according to the model influence level.

9 . The method of claim 8 , further comprising determining the one or more feature influence levels, determining the model influence level, and comparing at least one of the feature influence levels or the model influence level against the specified threshold in response to the event trigger, wherein the event trigger further comprises a user request to delete the individual-specific raw data instances.

10 . The method of claim 9 , further comprising:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, deleting the native AI model and the individual-specific raw data instances, and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, deleting the individual-specific raw data instances without deleting the native AI model.

11 . The method of claim 9 , further comprising:

determining whether the differential features associated with the individual are used in the native AI model; and

when the differential features are not used in the native AI model, deleting the individual-specific raw data instances without deleting the native AI model.

12 . The method of claim 9 , further comprising:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, deleting the native AI model and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, leaving the native AI model intact,

wherein the IHS comprises a multi-user IHS that administers the multi-user account.

13 . The method of claim 12 , wherein the native AI process further comprises a plurality of AI services, and the instructions are further executed to:

for each of a plurality of the AI services, comparing the feature influence levels or the model influence level against the specified threshold when the AI service has used the individual-specific raw data instances of the individual.

14 . The method of claim 8 , wherein the differential AI process is provided as a cloud-based service.

15 . A non-transitory memory storage device configured with program instructions stored thereon that, upon execution by one or more processors of an Information Handling System (IHS), cause the IHS to:

receive one or more individual-specific raw data instances associated with an individual, wherein the individual comprises a multi-user account;

receive a plurality of other raw data instances associated with individuals other than the individual;

receive a plurality of individual-specific resource data instances comprising measurements captured by sensors configured in the IHS;

perform a native artificial intelligence (AI) process to generate one or more native features in accordance to the received other raw data instances, the individual-specific raw data instances associated with the multi-user account, and the individual-specific resource data instances, generate a trained a native AI model from the one or more native features, the trained native AI model configured to generate one or more profile recommendations determined to optimize IHS performance and energy usage, wherein the one or more profile recommendations are determined by the trained native AI model as a function of IHS performance and energy usage feature input derived from telemetry data that includes measurements captured by the sensors configured in the IHS, wherein the one or more profile recommendations comprise one or more of: adjust one or more of a frame rate of a display resource, a refresh rate of a display resource, a computational frame rate of a GPU resource, an overclocking or underclocking level of a CPU resource, or a write optimized setting or a read optimized setting of a storage resource;

perform a differential AI process to generate one or more differential features in accordance to the received other raw data instances, and generate a differential model from the one or more differential features, wherein the differential features are generated without the individual-specific raw data instances; and

in response to an event trigger that indicates a request to remove influence of the individual from the trained native AI model, determine whether the influence of the individual on the trained native AI model was a high level of influence defined as at least 3 percent, based at least in part on:

compare the native features against the differential features to determine one or more feature influence levels of the individual-specific raw data instance on the native features;

compare the trained native AI model against the differential model to determine a model influence level of the individual-specific raw data instance on the trained native AI model;

compare at least one of the one or more feature influence levels, or the model influence level, to a specified threshold; and

in response to a determination that the influence of the individual on the trained native AI model was the high level of influence:

replace the native features and trained native AI model with the differential features and the differential model; and

delete the individual-specific raw data instances and the individual-specific resource data instances in accordance to the model influence level.

16 . The memory storage device of claim 15 , wherein the instructions are further executed to determine the one or more feature influence levels, determine the model influence level, and compare at least one of the feature influence levels or the model influence level against the specified threshold in response to the event trigger, wherein the event trigger further comprises a user request to delete the individual-specific raw data instances.

17 . The memory storage device of claim 16 , wherein the instructions are further executed to:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, delete the native AI model and the individual-specific raw data instances, and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, delete the individual-specific raw data instances without deletion of the native AI model.

18 . The memory storage device of claim 16 , wherein the instructions are further executed to:

determine whether the differential features associated with the individual are used in the native AI model; and

when the differential features are not used in the native AI model, delete the individual-specific raw data instances without deletion of the native AI model.

19 . The memory storage device of claim 16 , wherein the IHS comprises a multi-user IHS that administers the multi-user account, and the instructions are further executed to:

when at least one of the feature influence levels or the model influence level exceeds the specified threshold, delete the native AI model and schedule the native AI model to be re-trained; and

when at least one of the feature influence levels or the model influence level does not exceed the specified threshold, leave the native AI model intact.

20 . The memory storage device of claim 19 , wherein the native AI process further comprises a plurality of AI services, and the instructions are further executed to:

for each of the AI services, compare the feature influence levels or the model influence level against the specified threshold when the AI service has used the individual-specific raw data instances of the individual.