How it Works
Federated Analytics
What is Federated Analytics?
Federated analytics allows users to run distributed statistical analysis across multiple institutions without transferring raw data. Instead of centralizing datasets for exploration, each institution performs computations locally and only returns aggregated results. This helps researchers understand data characteristics, assess quality, and prepare analyses while maintaining institutional responsibility over sensitive information.
Learn more ➤How Federated Analytics Works
Running federated analytics requires a few setup steps on the institution side before a researcher can request statistics across the federation.
A data administrator adds a dataset to the institution and assigns it tags, making it identifiable by researchers across the federation.
Federated analytics must be explicitly enabled on the institution side by setting the corresponding flag in the configuration file.
The researcher requests statistics from the federation. Fed-BioMed first checks the cache and if results are already available, no new request is sent.
Each participating institution receives the request and computes statistics locally, returning only summable results and never raw data.
The researcher side collects responses from all institutions and aggregates them into the final statistics.
Privacy-Preserving Analytics
Fed-BioMed enforces privacy at every step of the analytics process, ensuring that individual institution data is never exposed.
Secure Aggregation
Data returned by institutions can be encrypted using secure aggregation. This can be enforced by the institution or requested by the researcher.
Aggregated Results Only
When secure aggregation is active, only the final aggregated statistics are accessible. Per-institution results are never visible.
Summable Statistics
To enable secure aggregation, institutions return summable values (for example, when 'mean' is requested, 'sum' and 'count' are sent instead).
Supported Data & Statistics
Currently, tabular data is supported. Available statistics include count, mean, and variance. Histogram support is under validation.