System and method for decentralized distributed model adaptation

1. An edge information handling system (IHS) manager, comprising:

a storage for storing:

labeled data associated with a use counter, wherein the labeled data is a portion of the labeled data stored in a central IHS; and

a vehicle counter;

an update pool; and

a processor programmed to:

receive, from an IHS, an update package,

wherein the update package comprises an unlabeled data update and an identifier corresponding to an inference model,

wherein the unlabeled data update is a result of an update calculation for the inference model performed by the IHS and using unlabeled data, and

wherein the unlabeled data is stored on the IHS and is not transmitted to the edge IHS manager;

update the inference module for the IHS using the labeled data and the update package;

determine, after the updating, whether the use counter of the labeled data has exceeded a current use threshold, wherein the current use threshold is based on the vehicle counter; and

in response to the use counter of the labeled data exceeding the current use threshold, initiate replacing of the labeled data with new labeled data from the central IHS,

wherein the central IHS, the edge IHS manager, and the IHS are operatively connected over a cloud network,

wherein the central IHS comprises all the labeled data associated with the use counter, and

wherein replacing of the labeled data with the new labeled data from the central IHS comprises:

moving a pointer associated with the labeled data to the update pool;

transmitting, based on determining that the update pool includes the pointer, a new labeled data request to a central IHS manager to retrieve the new labeled data;

receiving the new labeled data in response to the new labeled data request and deleting the labeled data associated with the pointer in the update pool from the storage; and

replacing the new labeled data in a portion of the storage previously storing the deleted labeled data to be used for updating subsequently received inference models.

2. The edge IHS manager of claim 1 , wherein the updating of the inference model using the labeled data comprises:

receiving an inference model update request specifying the inference model from a vehicle;

in response to receiving the inference model update request, updating the inference model using the labeled updating the vehicle counter based on the inference model update request to generate the current use threshold; and

in response to updating the inference model using the labeled data, updating the use counter of the labeled data by incrementing the user counter.

3. The edge IHS manager of claim 2 , wherein

the vehicle counter comprises a vehicle information storage and an active vehicle timestamp list, and

the updating of the vehicle counter based on the inference model update request to generate the current use threshold comprises:

parsing the inference model update request for a vehicle information of the vehicle;

storing the vehicle information in the vehicle information storage list;

updating the active vehicle timestamp list using the vehicle information to generate an updated active vehicle timestamp list;

parsing the updated active vehicle timestamp list to determine a number of active vehicles requesting inference model updates with the edge IHS manager; and

replacing a prior use threshold stored in the vehicle counter before the receipt of the inference model update request with the current use threshold.

4. The edge IHS manager of claim 3 , wherein

the storage further comprises a plurality of the labeled data each associated with a distinct one of the use counter, and

the processor is further programmed to randomly select one of the labeled data to update the inference model and update the use counter of the randomly selected one of the labeled data.

5. The edge IHS manager of claim 4 , wherein

each of the distinct ones of the use counter comprises a pointer associated with a pointer counter,

the pointer counter of a specific labeled data among the plurality of the labeled data is incremented whenever the specific labeled data is randomly selected to update the inference model.

6. The edge IHS manager of claim 1 , wherein the new labeled data is assigned the pointer in the update pool.

7. The edge IHS manager of claim 3 , wherein

the processor parses only a portion of the updated active vehicle timestamp list to determine the number of active vehicles, and

the portion of the updated active vehicle timestamp list parsed by the processor is determined based on a preset time window.

8. The edge IHS manager of claim 7 , wherein the portion of the updated active vehicle timestamp list is parsed starting from an oldest entry within the portion towards a newest entry within the portion.

9. The edge IHS manager of claim 4 , wherein the edge IHS manager stores less of the labeled data compared to the central IHS manager.

10. A method for updating labeled data stored in an edge information handling system (IHS) manager, the method comprising:

receiving, from an IHS, an update package,

wherein the update package comprises an unlabeled data update and an identifier corresponding to an inference model,

wherein the unlabeled data update is a result of an update calculation for the inference model performed by the IHS and using unlabeled data, and

wherein the unlabeled data is stored on the IHS and is not transmitted to the edge IHS manager;

updating an inference module for the IHS using the labeled data and the update package, wherein the labeled data is stored in a storage of the edge IHS manager;

determining, after the updating, whether the use counter of the labeled data has exceeded a current use threshold, wherein the current use threshold is based on a vehicle counter stored in the storage; and

in response to the use counter of the labeled data exceeding the current use threshold, initiating replacing of the labeled data with new labeled data from a central IHS,

wherein

the central IHS, the edge IHS manager, and the IHS are operatively connected over a cloud network,

wherein the central IHS comprises:

all the labeled data associated with the use counter; and

wherein replacing of the labeled data with the new labeled data from the central IHS comprises:

moving a pointer associated with the labeled data to the update pool;

transmitting, based on determining that the update pool includes the pointer, a new labeled data request to the central IHS manager to retrieve the new labeled data;

receiving the new labeled data in response to the new labeled data request and deleting the labeled data associated with the pointer in the update pool from the storage; and

replacing the new labeled data in a portion of the storage previously storing the deleted labeled data to be used for updating subsequently received inference models.

11. The method of claim 10 , wherein the updating of the inference model using the labeled data comprises:

receiving an inference model update request specifying the inference model from a vehicle;

in response to receiving the inference model update request, updating the inference model using the labeled data and updating the vehicle counter based on the inference model update request to generate the current use threshold; and

in response to updating the inference model using the labeled data, updating the use counter of the labeled data by incrementing the user counter.

12. The method of claim 11 , wherein

the vehicle counter comprises a vehicle information storage and an active vehicle timestamp list, and

the updating of the vehicle counter based on the inference model update request to generate the current use threshold comprises:

parsing the inference model update request for a vehicle information of the vehicle;

storing the vehicle information in the vehicle information storage list;

updating the active vehicle timestamp list using the vehicle information to generate an updated active vehicle timestamp list;

parsing the updated active vehicle timestamp list to determine a number of active vehicles requesting inference model updates with the edge IHS manager; and

replacing a prior use threshold stored in the vehicle counter before the receipt of the inference model update request with the current use threshold.

13. The method of claim 12 , wherein

the storage further comprises a plurality of the labeled data each associated with a distinct one of the use counter, and

the method further comprises randomly selecting one of the labeled data to update the inference model and update the use counter of the randomly selected one of the labeled data.

14. The method of claim 4 , wherein

each of the distinct ones of the use counter comprises a pointer associated with a pointer counter,

the pointer counter of a specific labeled data among the plurality of the labeled data is incremented whenever the specific labeled data is randomly selected to update the inference model.

15. A non-transitory computer readable medium (CRM) comprising computer readable program code, which when executed by a computer processor enables the computer processor to perform a method for updating labeled data stored in an edge information handling system (IHS) manager, the method comprising:

receiving, from an IHS, an update package,

wherein the update package comprises an unlabeled data update and an identifier corresponding to an inference model,

wherein the unlabeled data update is a result of an update calculation for the inference model performed by the IHS and using unlabeled data, and

wherein the unlabeled data is store on the IHS and is not transmitted to the edge IHS manager;

updating an inference module for the IHS using the labeled data and the update package, wherein the labeled data is stored in a storage of the edge IHS manger;

determining, after the updating, whether the use counter of the labeled data has exceeded a current use threshold, wherein the current use threshold is based on a vehicle counter stored in the storage; and

in response to the use counter of the labeled data exceeding the current use threshold, initiating replacing of the labeled data with new labeled data from a central IHS,

wherein

the central IHS, the edge IHS manager, and the IHS are operatively connecter over a cloud network,

wherein the central IHS comprises:

all the labeled data associated with the use counter; and

wherein replacing of the labeled data with new labeled data from the central IHS comprises:

moving a pointer associated with the labeled data to the update pool;

transmitting, based on determining that the update pool includes the pointer, a new labeled data request to the central IHS manager to retrieve the new labeled data;

receiving the new labeled data in response to the new labeled data request and deleting the labeled data associated with the pointer in the update pool from the storage; and

replacing the new labeled data in a portion of the storage previously storing the deleted labeled data to be used for updating subsequently received inference models.

16. The CRM of claim 15 , wherein the updating of the inference model using the labeled data comprises:

receiving an inference model update request specifying the inference model from a vehicle;

in response to receiving the inference model update request, updating the inference model using the labeled data and updating the vehicle counter based on the inference model update request to generate the current use threshold; and

in response to updating the inference model using the labeled data, updating the use counter of the labeled data by incrementing the user counter.

17. The CRM of claim 16 , wherein

the vehicle counter comprises a vehicle information storage and an active vehicle timestamp list, and

the updating of the vehicle counter based on the inference model update request to generate the current use threshold comprises:

parsing the inference model update request for a vehicle information of the vehicle;

storing the vehicle information in the vehicle information storage list;

updating the active vehicle timestamp list using the vehicle information to generate an updated active vehicle timestamp list;

parsing the updated active vehicle timestamp list to determine a number of active vehicles requesting inference model updates with the edge IHS manager; and

replacing a prior use threshold stored in the vehicle counter before the receipt of the inference model update request with the current use threshold.

18. The CRM of claim 17 , wherein

the storage further comprises a plurality of the labeled data each associated with a distinct one of the use counter, and

the method further comprises randomly selecting one of the labeled data to update the inference model and update the use counter of the randomly selected one of the labeled data.

19. The CRM of claim 18 , wherein

each of the distinct ones of the use counter comprises a pointer associated with a pointer counter,

the pointer counter of a specific labeled data among the plurality of the labeled data is incremented whenever the specific labeled data is randomly selected to update the inference model.