Non-iterative federated learning

1. A computer-implemented method comprising:

receiving, using a processor at a first site, local models from agents at remote sites, the local models including a first local model associated with a first generator model from a first remote site and a second local model associated with a second generator model from a second remote site, wherein the first local model and the first generator model were created at the first remote site, wherein the second local model and the second generator model were created at the second remote site;

preparing, by the processor, training data for a global model at the first site, the preparing training data comprising:

operating, using the processor, the first and second generator models to generate first and second synthetic datasets respectively for the first and second local models at the first site; and

operating, by the processor, the first local model having been created and received from the first remote site to generate a first output that labels the first synthetic dataset generated by the first generator model created and received from the first remote site in response to inputting the first synthetic dataset;

operating, by the processor, the second local model having been created and received from the second remote site to generate a second output that labels the second synthetic dataset generated by the second generator model created and received from the second remote site in response to inputting the second synthetic dataset, wherein the training data comprises the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset; and

training the global model with the training data comprising the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset.

2. The computer-implemented method of claim 1 , wherein the first and second synthetic datasets are statistically comparable to local datasets previously used to train the first and second local models.

3. The computer-implemented method of claim 1 further comprising transmitting the global model to each of the agents according to a type of model architecture requested by the agents.

4. The computer-implemented method of claim 1 , wherein the training the global model based on the first and second synthetic datasets and the first and second outputs enable unilaterally provisioning computing capabilities for fusion services.

5. A system comprising:

a memory having computer readable instructions; and

one or more processors at a first site for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:

receiving local models from agents at remote sites, the local models including a first local model associated with a first generator model from a first remote site and a second local model associated with a second generator model from a second remote site, wherein the first local model and the first generator model were created at the first remote site, wherein the second local model and the second generator model were created at the second remote site;

preparing training data for a global model at the first site, the preparing training data comprising:

operating the first and second generator models to generate first and second synthetic datasets respectively for the first and second local models at the first site;

operating the first local model having been created and received from the first remote site to generate a first output that labels the first synthetic dataset generated by the first generator model created and received from the first remote site in response to inputting the first synthetic dataset

operating the second local model having been created and received from the second remote site to generate a second output that labels the second synthetic dataset generated by the second generator model created and received from the second remote site in response to inputting the second synthetic dataset, wherein the training data comprises the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset; and

training the global model with the training data comprising the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset.

6. The system of claim 5 , wherein the first and second local models have been previously trained on a local dataset.

7. The system of claim 5 further comprising transmitting the global model to each of the agents according to a type of model architecture requested by the agents, wherein the first and second synthetic datasets are statistically comparable to local datasets previously used to train the first and second local models.

8. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor at a first site to cause the processor to perform operations comprising:

receiving local models from agents at remote sites, the local models including a first local model associated with a first generator model from a first remote site and a second local model associated with a second generator model from a second remote site, wherein the first local model and the first generator model were created at the first remote site, wherein the second local model and the second generator model were created at the second remote site;

preparing training data for a global model at the first site, the preparing training data comprising:

operating the first and second generator models to generate first and second synthetic datasets respectively for the first and second local models at the first site; and

operating the first local model having been created and received from the first remote site to generate a first output that labels the first synthetic dataset generated by the first generator model created and received from the first remote site in response to inputting the first synthetic dataset; and

operating the second local model having been created and received from the second remote site to generate a second output that labels the second synthetic dataset generated by the second generator model created and received from the second remote site in response to inputting the second synthetic dataset, wherein the training data comprises the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset; and

training the global model with the training data comprising the first output labeling the first synthetic dataset and the second output labeling the second synthetic dataset.