Granted Patent B2

US 12,518,157 · App. 17/435,468 · Granted Jan 6, 2026

Inventors: Tony Larsson (Upplands Väsby, SE); Johan Haraldson (Stockholm, SE)

Assignee: Telefonaktiebolaget LM Ericsson (Publ)

G06N3/08G06N3/044G06N20/00H04L41/16

Abstract

Claims (58)

1 . A computing device for dynamically configuring a network for training a machine learning model, the network comprising a server computing device and a plurality of client computing devices configured to perform training of the machine learning model, the computing device communicatively coupled to the network and comprising processing circuitry and a memory, the memory containing instructions executable by the processing circuitry to configure the computing device to:

select a number of the plurality of client computing devices to participate in training the machine learning model via federated learning;

determine a first value of an evaluation metric of the machine learning model based on the selected number of client computing devices;

indicate degraded performance of the machine learning model in response to comparing the first value of the evaluation metric to a threshold level;

determine the presence of an adjustment trigger;

adjust the number of client computing devices that participate in training the machine learning model in response to determining the presence of the adjustment trigger;

determine a second value of the evaluation metric based on the adjusted number of client computing devices;

set the number of client computing devices participating in training the machine learning model based on the second value of the evaluation metric; and

revert to the selected number of the plurality of client computing devices in response to comparing the second value of the evaluation metric to the first value of the evaluation metric.

2 . The computing device of claim 1 , further configured to:

determine the presence of an adjustment trigger by determining that the first value of the evaluation metric indicates performance of the machine learning model below a threshold level; and

adjust the number of client computing devices used to determine the value of the evaluation metric by increasing the number of client computing devices participating in training the machine learning model.

3 . The computing device of claim 2 , further configured to cause training of the machine learning model by the increased number of client computing devices before determining the second value of the evaluation metric.

4 . The computing device of claim 2 , further configured to set the number of client computing devices participating in training by one of:

reverting to the selected number of client computing devices if the second value of the evaluation metric is the same as or less than the first value of the evaluation metric;

maintaining the increased number of client computing devices if the second value of the evaluation metric indicates performance of the machine learning model above the threshold level; and

further increasing the number of client computing devices participating in training the machine learning model if the second value of the evaluation metric indicates performance of the machine learning model above the first value of the evaluation metric and below the threshold level.

5 . The computing device of claim 1 , further configured to determine the presence of an adjustment trigger by:

determining that the first value of the evaluation metric indicates performance of the machine learning model above a threshold level; and

determining that a predetermined period has passed since a previous adjustment to the number of client computing devices used to determine the value of the evaluation metric.

6 . The computing device of claim 5 , further configured to adjust the number of client computing devices used to determine the value of the evaluation metric by using a subset of the selected number of client computing devices to determine the value of the evaluation metric.

7 . The computing device of claim 5 , further configured to set the number of client computing devices participating in training by one of:

maintaining the selected number of client computing devices if the second value of the evaluation metric indicates performance of the machine learning model below the threshold level; and

decreasing the number of client computing devices participating in training to the number of client computing devices in the subset if the second value of the evaluation metric indicates performance of the machine learning model above the threshold level.

8 . The computing device of claim 5 , further configured to adjust the number of client computing devices used to determine the value of the evaluation metric by decreasing the number of client computing devices participating in training the machine learning model.

9 . The computing device of claim 8 , further configured to cause training of the machine learning model by the decreased number of client computing devices before determining the second value of the evaluation metric.

10 . The computing device of claim 8 , further configured to set the number of client computing devices participating in training by one of:

reverting to the selected number of client computing devices if the second value of the evaluation metric indicates performance of the machine learning model below the threshold level; and

maintaining the decreased number of client computing devices if the second value of the evaluation metric indicates performance of the machine learning model above the threshold level.

11 . The computing device of claim 1 , further configured to cause training of the machine learning model by the selected number of client computing devices before determining the first value of the evaluation metric.

12 . The computing device of claim 1 , wherein:

the network comprises a telecommunications network; and

the plurality of client computing devices comprises a plurality of access nodes of the telecommunications network.

13 . The computing device of claim 1 , further configured to:

determine a resource capacity of at least one of the client computing devices; and

select client computing devices to participate in training the machine learning model based on the determined resource capacities.

14 . The computing device of claim 13 , further configured to determine a resource capacity of at least one of the client computing devices by predicting the resource capacity at a time in the future at which the machine learning model will be trained.

15 . The computing device of claim 1 , wherein the computing device performing the method comprises the server computing device.

16 . The computing device of claim 1 , wherein the computing device performing the method comprises a plurality of computing devices.

17 . The computing device of claim 1 , further configured to determine the first value of the evaluation metric of the machine learning model periodically.

18 . A computer storage medium having stored thereon a computer program comprising instructions which, when executed on processing circuitry, causes the processing circuitry to perform method for dynamically configuring a network for training a machine learning model, the network comprising a server computing device and a plurality of client computing devices configured to perform training of the machine learning model, the method comprising:

selecting a number of the plurality of client computing devices to participate in training the machine learning model via federated learning;

determining a first value of an evaluation metric of the machine learning model based on the selected number of client computing devices;

indicating degraded performance of the machine learning model in response to comparing the first value of the evaluation metric to a threshold level;

determining the presence of an adjustment trigger;

adjusting the number of client computing devices that participate in training the machine learning model in response to determining the presence of the adjustment trigger;

determining a second value of the evaluation metric based on the adjusted number of client computing devices;

setting the number of client computing devices participating in training the machine learning model based on the second value of the evaluation metric; and

reverting to the selected number of the plurality of client computing devices in response to comparing the second value of the evaluation metric to the first value of the evaluation metric.

19 . A method for dynamically configuring a network for training a machine learning model, the network comprising a server computing device and a plurality of client computing devices configured to perform training of the machine learning model, the method performed at a computing device communicatively coupled to the network and comprising:

selecting a number of the plurality of client computing devices to participate in training the machine learning model via federated learning;

determining a first value of an evaluation metric of the machine learning model based on the selected number of client computing devices;

indicating degraded performance of the machine learning model in response to comparing the first value of the evaluation metric to a threshold level;

determining the presence of an adjustment trigger;

adjusting the number of client computing devices that participate in training the machine learning model in response to determining the presence of the adjustment trigger;

determining a second value of the evaluation metric based on the adjusted number of client computing devices;

setting the number of client computing devices participating in training the machine learning model based on the second value of the evaluation metric; and

reverting to the selected number of the plurality of client computing devices in response to comparing the second value of the evaluation metric to the first value of the evaluation metric.

Assignments (1)

Continuity (1)

References Cited (25)