Network nodes and methods for handling machine learning models in a communications network

1 . A method performed by a control network node in a communications network for handling machine learning, ML, models in the communications network comprising one or more network nodes, the method comprising:

obtaining, from a network node, an indication that the network node has experienced a change in the running of a first ML model in the network node, the indication being based on an increase in a loss value of the first ML model that indicates a change in input data to the first ML model;

determining to transmit to a network node in the communications network a pre-trained ML model, based on at least one of a signature and a loss value of the network node, the at least one of the signature and the loss value being related to ML modelling;

obtaining a signature of the network node from the network node, the signature being related to the first ML model and comprising a Gaussian model of all features the first ML model uses as input;

determining a similarity between the signature of the first ML model with one or more stored signatures by comparing the obtained signature of the first ML model with the stored signatures, the similarity being determined based on pairwise distance models;

selecting the pre-trained ML model having a signature most similar to the signature of the first ML model; and

transmitting the pre-trained ML model to the network node.

2 . The method according to claim 1 , further comprising:

determining that the network node is an added network node added to the communications network; and

the pre-trained ML model transmitted to the network node is selected based on the signature of the network node.

3 . The method according to claim 1 , wherein the indication is a received flag indication from another network node in the communications network.

4 . A method performed by a network node in a communications network for handling machine learning, ML, models in the communications network comprising one or more network nodes, the method comprising:

obtaining at least one of a signature and a loss value of the network node, the at least one of the signature and the loss value being related to a first ML model comprised in the network node and comprising a Gaussian model of all features the first ML model uses as input;

determining that the network node has experienced a change in the running of a first ML model in the network node based on an increase in a loss value of the first ML model that indicates a change in input data to the first ML model;

transmitting an indication to a control network node indicating that the network node has experienced the change in the running of the first ML model;

transmitting the obtained at least one of the signature and the loss value to a control network node in the communications network;

receiving a pre-trained ML model having a signature most similar to the signature of the first ML model, the pre-trained ML model having been selected based on a similarity between the signature of the first ML model with one or more signatures based on a comparison of the obtained signature of the first ML model with the signatures, the similarity being determined based on pairwise distance models; and

deploying the pre-trained ML model at the network node.

5 . The method according to claim 4 , wherein the indication is a flag indication.

6 . A control network node for handling machine learning, ML, models in a distributed ML model architecture, the control network node configured to:

obtain, from a network node, an indication that the network node has experienced a change in the running of a first ML model in the network node, the indication being based on an increase in a loss value of the first ML model that indicates a change in input data to the first ML model;

determine to transmit to a network node in a communications network a pre-trained ML model based on at least one of a signature and a loss value of the network node, the at least one of the signature and the loss value being related to ML modelling;

obtain a signature of the network node from the network node, the signature being related to the first ML model and comprising a Gaussian model of all features the first ML model uses as input;

determine a similarity between the signature of the first ML model with one or more stored signatures by comparing the obtained signature of the first ML model with the stored signatures, the similarity being determined based on pairwise distance models;

select the pre-trained ML model having a signature most similar to the signature of the first ML model; and

transmit the ML model to the network node.

7 . The control network node according to claim 6 , wherein the control network node is configured to determine that the network node is an added network node added to the communications network; and

The pre-trained ML model transmitted to the network node is based on the signature of the network node.

8 . The control network node according to claim 6 , wherein the indication is a received flag indication from another network node in the communications network.

9 . A network node for handling machine learning, ML, models in a ML model architecture, wherein the network node is configured to:

obtain at least one of a signature and a loss value of the network node, the at least one of the signature and the loss value being related to a first ML model comprised in the network node and comprising a Gaussian model of all features the first ML model uses as input;

determine that the network node has experienced a change in the running of a first ML model in the network node based on an increase in a loss value of the first ML model that indicates a change in input data to the first ML model;

transmit an indication to a control network node indicating that the network node has experienced the change in the running of the first ML model;

transmit the obtained at least one of the signature and the loss value to a control network node;

receive a pre-trained ML model having a signature most similar to the signature of the first ML model, the pre-trained ML model having been selected based on a similarity between the signature of the first ML model with one or more signatures based on a comparison of the obtained signature of the first ML model with the signatures, the similarity being determined based on pairwise distance models; and

deploy the pre-trained ML model at the network node.

10 . The network node according to claim 9 , wherein the indication is a flag indication.