Sharing Non-Reversible Data Statistics for Fast and Secure Federated Learning Native to Extreme Learning Machine.

Extreme Learning Machine (ELM) is a fast and simple machine learning method with an exact solution. This work presents its integration into conventional federated learning frameworks that preserve the method’s native property of a non-iterative and exact solution in data-distributed systems. Low computational complexity and data privacy features make federated ELM a good candidate for on-device computing in privacy-sensitive applications.