xTrial: Federated Learning on Health Data

The Problem

Federated learning promises collaborative model training without data sharing, but gradient updates can leak patient information through model inversion and membership inference attacks.

Research has demonstrated that gradient updates contain enough information to reconstruct individual training samples. In healthcare, this means patient records, diagnoses, and genomic data can be inferred from gradients alone.

Differential privacy adds noise that degrades model quality. Secure aggregation protocols add latency that makes large-scale federated training impractical across hospital networks with varying connectivity.

The Old Way

The PRIVATE.ME Solution

xTrial aggregates gradients on XorIDA shares via xCompute. No aggregation server sees raw gradients. The aggregation is information-theoretically secure — not noise-based, not computationally bounded.

Each hospital splits its gradient update into XorIDA shares distributed to independent aggregation nodes. Gradient summation runs on shares via xCompute’s XOR-gate-free addition. The aggregated gradient is reconstructed only after summation.

No noise is added — model quality is preserved. xLink M2M authenticates each hospital. HMAC-chained logs provide HIPAA-compliant training audit trails.

The New Way

How It Works

xTrial combines xCompute gradient aggregation with xLink hospital authentication and HMAC-chained training audit trails.

Use Cases

Integration

import { FedTrain } from '@private.me/xtrial';

const session = await FedTrain.create({
  hospitals: [hospitalA, hospitalB, hospitalC],
  modelConfig: modelDef,
  threshold: { k: 2, n: 3 }
});
const model = await session.trainRound(localGradients);

Viral Onboarding: < 2 Minute M2M Setup

One-Line Connection (SDK)

import { connect } from '@private.me/agent-sdk';

// Zero-config discovery
const trialSite = await connect('clinical-trial-site');

// Use immediately for trial data submission
const result = await trialSite.send({
  type: 'submit',
  data: { nctId: 'NCT04567890', siteId: 'SITE-42', patientData: gradients }
});

Network effects: When your CRO or research site already uses Xlink, connection is instant. When they don't, you send an invite (< 10 sec), they accept (< 60 sec), and all future connections are instant.

CLI Alternative

# Initialize once
xlink init

# Connect to trial coordinator
xlink connect trial-coordinator

# Invite a research site to join
xlink invite research-site-university

Zero-Downtime Migration

import { DualModeAdapter } from '@private.me/agent-sdk';

// Both Xlink AND your existing API key work
const adapter = new DualModeAdapter({
  xlink: trialAgent,
  fallback: { apiKey: process.env.FDA_TRIAL_API_KEY }
});

const result = await adapter.call('submitGradients', {
  nctId: 'NCT04567890',
  siteId: 'SITE-42',
  gradients: localGradients
});

// Track adoption progress
console.log(adapter.getMetrics());
// { xlinkPercentage: 73, xlinkCalls: 146, fallbackCalls: 54 }

As your trial sites and CROs adopt Xlink, your xlinkPercentage automatically increases. No code changes required.

Why Clinical Trials Need Viral M2M

• FDA compliance critical: 21 CFR Part 11 requires verifiable data integrity and audit trails. API key setup delays violate GCP submission deadlines.
• Multi-party coordination: Clinical trials involve sponsors, CROs, research sites, IRBs, data monitoring committees, and FDA. Manual key exchange across 8+ parties is untenable.
• Patient enrollment pressure: Trial sites compete for enrollment. Delays in data submission infrastructure cost weeks of patient recruitment.
• Network effects: Each sponsor's adoption reduces friction for their entire CRO network—sites connect instantly to multiple trials.

Each sponsor connects to ~3 CROs → exponential growth. Growth projection: Month 1 (100 sponsors) → Month 12 (74,185 sponsors) with VC = 1.2

Security Properties

Verifiable Data Protection

Every operation in this ACI produces a verifiable audit trail via xProve. HMAC-chained integrity proofs let auditors confirm that data was split, stored, and reconstructed correctly — without accessing the data itself.

Ready to deploy xTrial?

Talk to Ren, our AI sales engineer, or book a live demo with our team.

Ship Proofs, Not Source

xTrial generates cryptographic proofs of correct execution without exposing proprietary algorithms. Verify integrity using zero-knowledge proofs — no source code required.

Use Cases