9.1 What is Taiwan AI Labs Federated Learning Framework?

Taiwan AI Labs Federated Learning Framework includes two major parts:

1. Fed Machines.

2. Aggregator.

The Fed Machine is installed inside the institutes, and only needs to use the institute private network IP address to get the whole system work - no public IP address, no VPN needed, to reduce the risk of disposing the machine in the public domain. It could be setup just like other institute's internal servers, and only be reached by internal staffs. After a training plan is complete, participants could get their own local best-fit model at their own Fed Machine.

The Aggregator is where the aggregation algorithm run and combine all transmitted model weights. It needs to be set up with public IP (IPV4/V6) for communication with all Fed Machines. Trained model weights are aggregated there after each training round, and the final global model will be available there.

Note:

If you want to know the data flow of Taiwan AI Labs Federated Learning framework, please refer it here.

PS:

Below is a demonstration of Federated Learning progress, from creating the FL project, inviting participants, uploading datasets, initial AI model, weights, and then training the final result; by using Taiwan AI Labs Federated Learning Framework:

For more information of using Taiwan AI Labs Federated Learning Platform, please contact us directly.