Skip to main content

Federated ML Application at the Edge

The purpose of Federated ML is to provide Federated Learning Platform for data stored locally, improves accuracy in the edge computing. The high-level relationships between the functional domains is shown in the figure below:

Networking Domain: The Networking Domain uses a component called federated networking to create a safe cross-site networking among all the federated sits.

Computing and storage Domain: The Computing and Storage Domain provides local/distributed computing and storage abilities to support the federated computing mission.

Federated ML: The Domain called Federated ML is used to provide multi federated machine learning core components, such federated feature binning, federated feature selection, federated  algorithm (LR, BOOST,NN)  and etc.

FATE-FLOW: The Domain called FATE-FLOW is used to schedule federated machine learning task, parse the flow DAG and manage the lifecycle of a federated machine learning task.

FATE-SERVING: The Domain called FATE-SERVING is used to generate and provide an online interface for the federated model and manage the federated model.

FATE-BOARD: The Domain called FATE-BOARD is used to visualize the federated machine learning task, monitor the federated model status and manage the logs.

The Federated ML Reference Architecture is shown in the figure below. For the full description of the Federated ML Reference Architecture please refer to the Federated Architecture Document.

Key features and implementations in Akraino Release 4:

  • FATE first unsupervised learning algorithm: Hetero KMeans
  • Add Data Split module: splitting data into train, validate, and test sets inside FATE modeling workflow
  • Add DataStatistic module: compute min/max, mean, median, skewness, kurtosis, coefficient of variance, percentile, etc.
  • Add PSI module for computing population stability index



For more information visit our wiki.