Unity Network
  • Unity Network
    • Introduction
    • Unity Network FL Architecture
    • Data Security
    • Case Study
      • Flashback Overview
      • Registration
      • Integration
      • Model Training
      • Objectives and Outcomes
  • Unity Network SDK
    • Key Features
    • Registration
    • Node Setup
    • SDK Integration
      • Wallet Module
      • Node Module
  • Permissions Management
    • User Permissions
  • Model Training
    • Organization registration
    • Model Training Requests
    • Secure Training and Updates
      • Model Training
        • Model Loading and Initialization
        • Data Loading
        • Training
      • Secure Transmission of Encrypted Updates
        • ECDH Key Exchange for Secure Encryption Key Generation
        • Encrypt and Transmit Model Updates
        • Secure Aggregation and Decryption at Central Server
        • Distribute Updated Model and Continue Training
      • Sharing Model Updates with the Model Owner and Verifying Authenticity of Training
        • Construct the Merkle Tree and Commit to the Merkle Root
        • Log Hashes of Accessed Dataset Chunks During Training
        • Transmission of Model Updates, Merkle Proofs, and Hash Log to the Model Owner
        • Verification by the Model Owner
  • Training rewards
    • Incentivization Process
Powered by GitBook
On this page
  1. Model Training
  2. Secure Training and Updates

Sharing Model Updates with the Model Owner and Verifying Authenticity of Training

Objective: To securely share the final, aggregated model updates with the model owner while verifying that training was conducted on the correct dataset. Using Merkle Trees with Hash Logging enables data integrity verification by ensuring that training accessed only the committed dataset.

Process Overview

  1. Merkle Tree Construction and Commitment: Each node constructs a Merkle Tree for its dataset and commits to the Merkle root as a cryptographic fingerprint of the data.

  2. Hash Logging During Training: Nodes log the hashes of dataset chunks accessed during training. These logs ensure that the model trained on the correct dataset.

  3. Verification of Hash Logs and Merkle Proofs: The model owner verifies the Merkle root, Merkle proofs, and hash logs to confirm that training was conducted on the committed dataset.

PreviousDistribute Updated Model and Continue TrainingNextConstruct the Merkle Tree and Commit to the Merkle Root

Last updated 7 months ago