Distributed Secure Model Training Framework for Privacy Protection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current data cooperation technologies face challenges such as data privacy leakage risks, lack of trust between parties, inefficiencies in centralized modeling solutions, and inaccuracies in machine learning models due to homomorphic encryption and differential privacy methods, which result in slow computational speeds and high communication costs.

Innovation Solution

A distributed multi-party secure model training framework using secret sharing and an event-driven interactive secure modeling procedure, where secure logistic regression models are iteratively updated by secure computation nodes that keep their training data secret, using random numbers for encryption and obfuscation, and an independent third-party random number service to prevent information leakage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If centralized data aggregation and traditional machine learning modeling is used, then model training can be uniformly performed, but data privacy leakage risks increase and trust requirements between parties become more stringent

Engineering Contradiction:
Improvedata privacy protectionVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system segments the centralized modeling process into distributed secure computation nodes, each performing local computations on encrypted data. This segmentation eliminates the need for centralized data aggregation while maintaining modeling capability, thereby protecting data privacy without requiring complex trust mechanisms between parties

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a random number provider as an intermediary that generates encryption keys and random numbers distributed to secure computation nodes. This intermediary enables encrypted data processing without requiring direct trust between participating parties, resolving the privacy protection versus system complexity contradiction

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If homomorphic encryption and differential privacy methods are applied, then data privacy is protected, but computational speed decreases and communication costs increase

Engineering Contradiction:
Improveprivacy protectionVSAvoidcomputational efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system applies different privacy protection mechanisms at different stages: local encryption at secure computation nodes using random numbers generated by a third party, rather than applying heavy homomorphic encryption throughout the entire pipeline. This localized approach maintains privacy while improving computational efficiency

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent uses disposable random numbers generated by a trusted third party for encryption, replacing expensive and computationally intensive homomorphic encryption schemes. These random numbers are used once and then discarded, providing adequate privacy protection with minimal computational overhead and improved productivity

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Reliability

If secret sharing and distributed secure computation is implemented, then data privacy is enhanced, but communication costs and system complexity increase

Engineering Contradiction:
Improveprivacy protectionVSAvoidcommunication cost
Core Design Contradiction:
ReliabilityVSLoss of energy

Solution Approach 1:

Each secure computation node performs self-service by locally encrypting its data using random numbers from the third party and independently contributing to the distributed model training. This eliminates the need for complex inter-node communication and coordination, reducing communication costs while maintaining enhanced privacy protection

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3825887B1A distributed multi-party security model training framework for privacy protection
Publication Date: 2022.03.23 ADVANCED NEW TECHNOLOGIES CO LTD
  • EP3825887B1 patent drawingFigure 1
  • EP3825887B1 patent drawingFigure 2A
  • EP3825887B1 patent drawingFigure 2B

AI summary

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a multi-party secure logistic regression model (SLRM). One of the methods includes receiving, at a plurality of secure computation nodes (SCNs), a plurality of random numbers from a random number provider; encrypting, at each SCN, data stored at the SCN using the received random numbers; iteratively updating a secure logistic regression model (SLRM) by using the encrypted data from each SCN; and after iteratively updating the SLRM, outputting a result of the SLRM, wherein the result is configured to enable a service to be performed by each SCN.