FAQs

What is PrivateAIM and FLAME?

PrivateAIM is the main project from MII consortium and Flame is the name of the platform.

What are the key components of the FLAME platform?

The major components are the Hub and the Node, both offer a set of services which interconnect with each other.

What is the difference between Hub and Node?

The HUB and NODE are two separate components of FLAME with different roles:

HUB (central platform) services:

• Acts as the central control point for the entire FLAME system.
• Manages authentication, authorization, and user access across all participating organizations.
• Organizes institutions into Realms.
• Analysis Management: Receives analysis code, builds Docker images, distributes to nodes, and receives results.
• Message Broker: Routes communications between nodes and users.
• Storage Service: Central repository to load analysis scripts.

NODE (local installation):

A node is installed locally at each participating hospital/clinic/university. Focuses on data management and analysis execution on the stored data defined from a FHIR server or from tabular files.

Node services:

• Data Store Setup: Manages FHIR and S3 data storage locally.
• Analysis Execution: Runs approved analyses on local data
• Node Aggregation: One node per analysis acts as "aggregator" to combine results from all node participants
• Monitoring: Local admin can view analysis execution status and logs.

What is a Realm in FLAME?

A Realm is an administrative namespace that represents one organization within the Hub.

Keypoints:

• Is assigned one Realm per organization
• Created by the main Hub administrator and assigned to administrators from each organisation.
• Each administrator manage their own Realm's:
  • Users and user accounts
  • Roles and permissions
  • Identity providers (for authentication)
  • Registered nodes (local installations)

What is the purpose of a Node as Aggregator?

An Aggregator is a specialized node that combines results from multiple analyzer nodes into a final aggregated result in a federated analysis.

Key Purposes:

1. Combine Results from all Analyzers: Receives analysis results from every participating analyzer node and implements the aggregation_method() that defines HOW to combine those results.
2. Define Aggregation Logic. The aggregator implements the business logic for combining results, such as:
   • Summing,
   • Averaging,
   • Weighted Combination: Combine results with different weights based on data size
3. Control iterative analysis. The aggregator decides when to stop or continue multi-iteration analyses
4. Send final results to the HUB. After aggregation is complete, the aggregator sends the final aggregated result to the Hub 🎉.

Further questions

What is Federated Learning?

Federated learning (FL) is generally defined as a machine-learning pattern in which multiple participants collaboratively train a shared model while keeping their training data at the originating devices or organizations McMahan et al.