Federated Learning Service

In this page you can have a detailed description of the Federated Learning. Follow the instructions to experiment with the service through the Secured Innohub.

Federated learning is a collaborative approach where a small number of trusted organizations, jointly train a machine learning model without sharing their sensitive data. For example, multiple hospitals may each hold electronic health records of their patients, but due to strict privacy regulations like GDPR, they cannot pool this data into a central server. Instead, each hospital trains the model locally on its own data, and only the model updates (not the raw data) are shared and combined to build a stronger global model. This way, the hospitals benefit from each other’s knowledge — for instance, improving disease prediction models by learning from a wider and more diverse patient population — while ensuring that no private medical records ever leave their premises.

Description

The Innohub Federated Learning service eases the orchestration of the collaborative training via a GUI to configure the Innohub FL Tool

In the SECURED InnoHUB, we provide a federated learning service built on top of Flower, an open-source FL framework, and packaged in a Docker container for easy deployment. Unlike classical federated learning setups that rely on a separate central aggregation server, our implementation uses a client–server hybrid approach: one of the participating clients also takes the role of the server. This means the institute not only trains its own model locally but also orchestrates the aggregation of updates from all collaborating partners. This design reduces infrastructure overhead and makes it easier for organizations to set up and manage their own federations.
Through the InnoHUB FL service, clients who want to initiate a collaboration can configure a new federation by specifying key parameters — such as the machine learning model type, hyperparameters, and expected data formats — and by inviting other institutes to join. Multiple federations can be created with different configurations, clients, and datasets, allowing parallel experiments to run seamlessly. The service generates the configuration file needed for the Docker-based Flower setup, which then executes the federation. This provides a user-friendly way for organizations to launch and manage secure, collaborative training without needing deep technical expertise in federated learning frameworks.

Requirements

To run this federated learning setup within the SECURED InnoHUB, each participating client needs to meet some basic hardware and software requirements. On the hardware side, a modern workstation or server with sufficient CPU power, memory (e.g., 16–32 GB RAM), and optionally a GPU for faster training is recommended, though exact specifications depend on the chosen model and dataset size. On the software side, Docker is the main requirement, as it encapsulates the entire Flower-based FL framework and its dependencies, minimizing installation complexity. Each client must also configure its IP address in the InnoHUB profile, where multiple IPs can be defined to allocate different federated experiments across different machines. This flexible setup allows institutes to dedicate specific resources to specific federations, ensuring scalability and smooth parallel execution.

Experimentation

Please Login in order to have access to the installation material.