Federated learning is a machine learning technique that trains a model over several dispersed nodes or hosts, as the name suggests. Each node utilises its own training data. If the model parameters are shared between nodes rather than the raw data, the data can be kept private.

Due to privacy concerns, obtaining training data to design and evolve machine learning models is increasingly being questioned, and federated learning can help alleviate some of these issues.

The Chinese e-commerce behemoth, Alibaba, has created a federated learning platform that allows machine learning algorithms to be constructed without sharing training data, as reported by MARTECHPOST.COM.

The source code for FederatedScope was released on GitHub under the Apache 2.0 license. The platform is a comprehensive federated learning platform that allows for flexible customization for many machine learning applications in academia and industry.

According to Alibaba, FederatedScope has an event-driven architecture and offers several tools, including a collection of benchmark datasets, well-known model architectures, sample federated learning algorithms, and automatic tuning mechanisms.

Developers can use these capabilities to create and configure task-specific federated learning systems in computer vision, natural language processing, speech recognition, graph learning, and recommendation.

FederatedScope also provides privacy protection through differential privacy and multi-party computing to suit various privacy requirements.