Federated Learning

Federated Learning

Privacy-preserving distributed machine learning directly in the database.

Overview

HeliosDB Federated Learning enables:

• Distributed ML training across nodes without sharing raw data
• Privacy-preserving model aggregation
• Differential privacy guarantees
• Integration with database queries

Quick Start

-- Create a federated learning job
CREATE FEDERATED LEARNING JOB fraud_detection
  MODEL TYPE 'logistic_regression'
  USING (SELECT features, label FROM transactions)
  WITH (
    rounds = 10,
    local_epochs = 5,
    privacy_budget = 1.0
  );

-- Start training
START FEDERATED JOB fraud_detection;

-- Check training status
SELECT * FROM helios_federated_jobs WHERE name = 'fraud_detection';

Key Features

Feature	Description
Privacy-Preserving	Data never leaves local nodes
Differential Privacy	Configurable privacy guarantees
Secure Aggregation	Encrypted gradient aggregation
Model Types	Logistic regression, neural networks, XGBoost
Auto-Scaling	Automatic participant management

Documentation

Document	Description
USER_GUIDE.md	Complete user guide

Related

• ML Integration: /docs/guides/user/ADVANCED_ML_INTEGRATION_GUIDE.md
• GPU Acceleration: /docs/guides/user/GPU_ACCELERATION_GUIDE.md

---

Status: Production Ready Version: v7.0