Reinforced federated learning utilizing multiple specialized machine learning agents

1 . A method for developing a machine-learning model, the method comprising:

receiving, at a central node, at least one episode comprising a plurality of changes of an environment;

analysing the episode to extract a plurality of observations and grouping the observations from among the plurality of observations into a plurality of groups of similar observations;

training a first machine learning agent using a first group of similar observations from among the plurality of groups of similar observations;

training a second machine learning agent using a second group of similar observations from among the plurality of groups of similar observations, wherein the second group of similar observations is different to the first group of similar observations; and

obtaining, at the central node, a central machine-learning model based on an output from at least one of the trained first machine learning agent and the trained second machine learning agent.

2 . The method of claim 1 , wherein the central node is further configured to trigger a generation of a plurality of machine learning agents, the plurality of machine learning agents comprising the first machine learning agent and the second machine learning agent.

3 . The method of claim 2 , wherein the plurality of machine learning agents analyse the episode to extract observations.

4 . The method of claim 3 , wherein:

each observation is a tuple of the form (s,a,r,s′) where s is the state of the environment prior to performing an action, a is an action, r is the reward that the machine learning agent analysing the episode would receive based on action a, and s′ is the state of the environment following action a; and

the observations are stored in an observation pool.

5 . The method of claim 1 , wherein the observations are grouped according to similarity using machine reasoning techniques.

6 . The method of claim 5 wherein the machine reasoning techniques comprise logic based mechanisms.

7 . The method of claim 1 , wherein the central node obtains a first machine learning model from the trained first machine learning agent and a second machine learning model from the trained second machine learning agent.

8 . The method of claim 7 , wherein the central node combines the first machine learning model and the second machine learning model to obtain the central machine learning model.

9 . The method of claim 8 , wherein the first machine learning model and the second machine learning model are combined by averaging the models.

10 . The method of claim 9 , wherein the first machine learning model and the second machine learning model are combined by averaging the models using a weighted average, a weight of each model being based on the total reward obtained by the model during training of the respective model using a respective group of similar observations.

11 . The method of claim 7 , wherein the first machine learning model and the second machine learning model are combined by selecting portions of the first machine learning model and portions of the second machine learning model.

12 . The method of claim 11 , wherein the selection of portions is controlled by a selecting machine learning model.

13 . The method of claim 1 , wherein the first machine learning agent and second machine learning agent are trained using reinforcement learning.

14 . The method of claim 1 , wherein the environment is a 3rd Generation Partnership Project, 3GPP, network, and wherein the observations are grouped based on Key Performance Indicator, KPI, degradation metrics.

15 . The method of claim 14 , wherein:

the first group of similar observations are observations indicating KPI deterioration and the second group of similar observations are observations indicating KPI improvement; and

the method further comprises training a third machine learning agent using a third group of similar observations, the third group of similar observations indicating KPI stability.

16 . The method of claim 14 , wherein the central machine learning model is used to suggest a network modification to help address KPI degradation.

17 . The method of claim 16 , wherein the network modification comprises altering the downlink power of one or more base stations within the network and/or switching a multiplexing scheme of one or more users of the network.

18 . A central node configured to develop a machine learning model, the central node comprising processing circuitry and a memory containing instructions executable by the processing circuitry, whereby the central node is operable to:

receive at least one episode comprising a plurality of changes of an environment;

initiate an analysis of the episode to extract a plurality of observations and group the observations from among the plurality of observations into a plurality of groups of similar observations;

initiate a training of a first machine learning agent using a first group of similar observations from among the plurality of groups of similar observations;

initiate a training of a second machine learning agent using a second group of similar observations from among the plurality of groups of similar observations, wherein the second group of similar observations is different to the first group of similar observations; and

obtain a central machine-learning model based on an output from at least one of the trained first machine learning agent and the trained second machine learning agent.

19 . The central node of claim 18 , further configured to trigger a generation of a plurality of machine learning agents, the plurality of machine learning agents comprising the first machine learning agent and the second machine learning agent.

20 . The central node of claim 19 , wherein the plurality of machine learning agents analyse the episode to extract observations.