The most popular and comprehensive Open Source ECM platform
Federated Learning is a way to train machine learning algorithms on data that is partitioned and distributed across many servers that each train on a subset of all data, then send their trained algorithm parameters back to a central server that aggregates the parameters and from that, creates a new composite algorithm.
An obvious benefit of federated learning is speed. There are more servers working in parallel on the problem, so the time to build a complete trained algorithm is shorter. But a bigger benefit is that the approach helps to better anonymize the data. Rather than pooling all the data onto a single server, the central server doesn’t have a copy of the original proprietary data, only the results of the training. This has been identified as a way to achieve better data privacy.
The idea of federated machine learning is being used by some researchers now for interacting with health care data. Algorithms can be trained without storing all the data in a centralized location. This allows them to take advantage of machine learning insights while still complying with the privacy requirements of HIPAA.