IBM Federated Learning
Federated Learning provides the tools for multiple remote parties to collaboratively train a single machine learning model without sharing data. Each party trains a local model with a private data set. Only the local model is sent to the aggregator to improve the quality of the global model that benefits all parties.
Required service
Watson Machine Learning. You must install the Watson Machine Learning service instance in Cloud Pak for Data as a Service to use Federated Learning. Federated Learning is available when you install Watson
Machine Learning.
Data format
Any data format including but not limited to CSV files, JSON files, and databases for PostgreSQL.
How Federated Learning works
Watch this overview video to learn the basic concepts and elements of a Federated Learning experiment. Learn how you can apply the tools for your company's analytics enhancements.
This video provides a visual method to learn the concepts and tasks in this documentation.
An example for using Federated Learning is when an aviation alliance wants to model how a global pandemic impacts airline delays. Each participating party in the federation can use their data to train a common model without ever moving or sharing their data. They can do so either in application silos or any other scenario where regulatory or pragmatic considerations prevent users from sharing data. The resulting model benefits each member of the alliance with improved business insights while lowering risk from data migration and privacy issues.
As the following graphic illustrates, parties can be geographically distributed and run on different platforms. One member might work on a physical server with a relational database in North America, another member works with a virtual machine and Cloud Object Storage in Europe, and the last member might work in Australia with their laptop and a CSV file.
Why use IBM Federated Learning
IBM Federated Learning has a wide range of applications across many enterprise industries. Federated Learning:
- Enables sites with large volumes of data to be collected, cleaned, and trained on an enterprise scale without migration.
- Accommodates for the differences in data format, quality, and constraints.
- Complies with data privacy and security while training models with different data sources.
Learn more
- Federated Learning tutorials and samples
- Getting started
- Frameworks, fusion methods, and Python versions
- Creating a Federated Learning experiment
- Applying encryption
- Limitations and troubleshooting
Parent topic: Analyzing data and building models