Federated & secure computation

Learn from data
you can't move.

Hospital and research data is often too sensitive to share. So we bring the computation to the data instead, with federated learning, federated analytics and secure multi-party computation.

Talk to us → What is federated analytics?
Site A Raw data stays Site B Raw data stays Site C Raw data stays results only Combined model or answer Site A Site B Site C results only Combined answer
Fig. 1 The computation travels; the data stays

01 Three methods

Different questions, same principle.

A

Federated learning

Train one shared AI model across many sites. Each site trains on its own data, and only the model updates are combined. The model learns from everyone's data without anyone handing it over.

B

Federated analytics

Answer statistical questions across many sites, such as counts, averages, distributions and group comparisons, without pooling the raw data. How it works →

C

Secure multi-party computation

Several parties compute a joint result while each party's inputs stay hidden, even from the others. Data is split into encrypted shares, so no single party can reconstruct anyone else's data.

02 Federated analytics, explained

Not every question needs a model.

Many research questions are about measuring, not predicting. How many patients fit a profile? How does a signal differ between two groups? Federated analytics answers these across many sites at once. Each site runs the calculation on its own data and sends back only a summary, never individual records.

Example

A research team wants to compare average EEG alpha power between two patient groups across five hospitals. Each hospital computes its own group averages and counts locally. Only those summaries are combined into the overall result. The team gets the answer; no one sees a single patient's recording.

Step 1 Ask The question goes to every site Step 2 Compute locally Each site runs it on its own data Step 3 Share summaries Only aggregates leave the site Step 4 Combined answer As if the data were in one place Step 1 Ask every site Step 2 Compute locally Step 3 Share summaries only Step 4 Combined answer
Fig. 2 How a federated analytics query runs

03 Side by side

Which one do you need?

Federated learning Federated analytics Secure MPC
Produces A trained model Statistics and answers A joint result
What leaves each site Model updates Aggregate summaries Encrypted shares
Best for Prediction and classification Cohort counts, group comparisons, distributions Computing together when no party should see another's inputs

The three can be combined. For example, secure multi-party computation can protect the updates or summaries that federated learning and analytics produce.

Data you can't share? Let's talk.

Tell us what you want to learn and where the data lives. We'll show you how to get the answer without moving it.

Contact us →