System and method for training federated learning model

A system for training a federated learning model in a federated learning network comprising a server and a plurality of clients, the server maintaining a global federated learning model, the plurality of clients separately maintaining decentralized data sources, the data sources separately storing training datasets for the global federated learning model, the system comprising:

• a memory having stored thereon executable instructions;
• a processor circuitry in communication with the memory, the processor circuitry when executing the instructions configured to:
• receive the global federated learning model from the server via the client, the client controlling remote computing resources;
• identify a spare computing instance from the remote computing resources;
• determine a threshold training load for training the global federated learning model based on a training load assigned to the client, the threshold training load being a subset of the assigned training load, wherein the training load assigned to the client is n epochs and the threshold training load is at least 50% of the η epochs associated to the training load assigned to the client;
• in response to a processing capacity of the spare computing instance being sufficient to process the threshold training load, offload the threshold training load to the spare computing instance;
• and
• train the global federated learning model on the spare computing instance with the training dataset stored in a data source maintained by the client,
• in response to the spare computing instance completing the threshold training load but having insufficient processing capacity to complete the assigned training load, transmit a partially trained model update to the server.

A product for training a federated learning model in a federated learning network comprising a server and a plurality of clients, the server maintaining a global federated learning model, the plurality of clients separately maintaining decentralized data sources, the data sources separately storing training datasets for the global federated learning model, the product comprising:

• non-transitory machine-readable media;
• and
• instructions stored on the machine-readable media, the instructions configured to, when executed, cause a processor circuitry to:
• receive the global federated learning model from the server via the client, the client controlling remote computing resources;
• identify a spare computing instance from the remote computing resources;
• determine a threshold training load for training the global federated learning model based on a training load assigned to the client, the threshold training load being a subset of the assigned training load, wherein the training load assigned to the client is n epochs and the threshold training load is at least 50% of the n epochs associated to the training load assigned to the client;
• in response to a processing capacity of the spare computing instance being sufficient to process the threshold training load, offload the threshold training load to the spare computing instance;
• and
• training the global federated learning model on the spare computing instance with the training dataset stored in a data source maintained by the client,
• in response to the spare computing instance completing the threshold training load but having insufficient processing capacity to complete the assigned training load, transmit a partially trained model update to the server.