Apparatuses and methods for collaborative learning
A server ( 12 ) comprising means ( 123 ) for: generating at least one updated set of model parameters based on at least one current set of model parameters received from at least one apparatus ( 100 - n ), the at least one updated set of model parameters being parameters of a machine learning model and being an update of the at least one current set of model parameters; sending the at least one updated set of model parameters to at least one apparatus ( 100 - n ).
1. A server comprising: at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the server at least to perform:
generating at least one updated set of model parameters based on at least one current set of model parameters received from at least one apparatus, the at least one updated set of model parameters being parameters of a machine learning model and an update of the at least one current set of model parameters, and the at least one updated set of model parameters being generated according to a Gaussian model;
sending the at least one updated set of model parameters to at least one apparatus and enabling prevention of tampering with inference data.
2. The server of claim 1 configured to perform generating the at least one updated set of model parameters further configured to perform a centralized processing based on two or more current sets of model parameters received from two or more corresponding apparatuses.
3. The server of claim 2 further configured to group the two or more apparatuses into one or more clusters, wherein the server are configured to generate at least one updated set of model parameters per cluster of apparatuses.
4. The server of claim 2 , further configured to generate the at least one updated set of model parameters by applying a weighted average to the two or more current sets of model parameters using two or more weighting coefficients, wherein the server is further configured to generate the weighting coefficient applied to one set of current model parameters received from one apparatus based on one or more of an accuracy value, analytics data, and a number of training samples received from the one apparatus, wherein the accuracy value represents an accuracy of the machine learning model configured with the one set of current model parameters and trained based on the training samples, wherein the analytics data are obtained from inference data generated at output of two or more trained machine learning models configured respectively with the two or more current sets of model parameters.
5. The server of claim 3 , further configured to generate the at least one updated set of model parameters by applying a weighted average to the two or more current sets of model parameters using two or more weighting coefficients, wherein the server is further configured to generate the weighting coefficient applied to one set of current model parameters received from one apparatus based on one or more of an accuracy value, analytics data, and a number of training samples received from the one apparatus, wherein the accuracy value represents an accuracy of the machine learning model configured with the one set of current model parameters and trained based on the training samples, wherein the analytics data are obtained from inference data generated at output of two or more trained machine learning models configured respectively with the two or more current sets of model parameters.
6. The server of claim 1 configure to generate an updated accuracy value representing an accuracy of the machine learning model configured with the at least one updated set of model parameters, further configured to determine a time to send the at least one updated set of model parameters to the at least one apparatus based on one or more of the updated accuracy value, a number of the at least one current set of model parameters, and a model update timer.
7. The server of claim 1 configured to generate the at least one updated set of model parameters for at least one machine learning use-case, further configured to send the at least one updated set of model parameters to at least one apparatus in response to a subscription request sent by the at least one apparatus to receive one or more updated sets of model parameters for the machine learning use-case.
8. An apparatus comprising: at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform:
sending a current set of model parameters to a server, the current set of model parameters being parameters of a trained machine learning model;
configuring the trained machine learning model with an updated set of model parameters received from the server and enabling prevention of tampering with inference data, the updated set of model parameters being an update of the current set of model parameters, and the at least one updated set of model parameters being generated according to a Gaussian model.
9. The apparatus of claim 8 , further configured to:
obtain an accuracy value representing an accuracy of the trained machine learning model configured with the current set of model parameters;
determine a time for sending the current set of model parameters to the server based on one or more of the accuracy value and a model update timer.
10. The apparatus of claim 9 configured to obtain the current set of model parameters from a training of a machine learning model performed based on training samples, further configured to send to the server at least one of: the accuracy value, analytics data, one or more formats of input data, a feature extraction algorithm, one or more data pre-processing techniques, and a number of the training samples used for training the machine learning model, wherein the analytics data are obtained from inference data generated by running the trained machine learning model configured with the current set of model parameters using the input data.
11. The apparatus of claim 8 , wherein the machine learning model is implemented for a machine learning use-case, the at least one processor and the at least one memory storing instructions, when executed by the at least one processor, are further configured to cause the apparatus to send a subscription request to receive from the server at least one updated set of model parameters for the machine learning use-case.
12. The apparatus of claim 8 , wherein the apparatus is a distributed self-organizing network apparatus operable at a wireless edge of a wireless communication system, wherein the at least one processor and the at least one memory storing instructions, when executed by the at least one processor, are further configured to cause the apparatus to perform one or more operations by running the trained machine learning model configured with the updated set of model parameters, an operation being a radio access network optimization operation at the wireless communication system.
13. The apparatus of claim 11 , wherein the subscription request comprises an identifier of a use-case specific classification algorithm.
14. The server of claim 7 , wherein the subscription request comprises an identifier of a use-case specific classification algorithm.
15. The apparatus of claim 8 , wherein the apparatus is an Internet of Things device.