Vertical Federated Learning Noise Matrix Encryption

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Vertical federated learning faces inefficiencies due to high computational overhead from homomorphic encryption, affecting the performance of machine learning algorithms and data privacy in multi-party data sharing scenarios.

Innovation Solution

A method that involves calculating a noise matrix and using it to encrypt residual vectors, allowing for noise addition processing without exposing private data, thereby improving efficiency and ensuring data privacy by encrypting residual vectors sent between data parties.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If homomorphic encryption is used to encrypt data before output from trust domain, then data privacy is protected, but computational overhead increases significantly

Engineering Contradiction:

Improvedata privacyVSAvoidvertical federated learning efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent replaces expensive homomorphic encryption with a cheaper encryption scheme based on matrix operations and noise addition. The encryption uses random matrices and additive noise that can be efficiently generated and discarded after use, significantly reducing computational overhead while maintaining privacy protection for data in vertical federated learning.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent changes the encryption parameters from complex homomorphic encryption operations to simpler matrix multiplication and addition operations. By transforming the encryption mechanism into operations involving random matrices and controllable noise levels, the system achieves comparable security with much lower computational cost.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If homomorphic encryption is applied to protect sensitive data, then data security is ensured, but machine learning algorithm performance deteriorates

Engineering Contradiction:

Improvedata securityVSAvoidcomputational power consumption

Core Design Contradiction:

ReliabilityVSPower

Solution Approach 1:

The patent substitutes the mechanical complexity of homomorphic encryption operations with a different encryption mechanism based on linear algebra operations (matrix multiplication and addition). This substitution replaces computationally intensive cryptographic operations with more efficient matrix operations that consume less computational power while achieving the same security objective.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If data is encrypted before output from trust domain, then privacy is preserved, but computational efficiency decreases

Engineering Contradiction:

Improveprivacy preservationVSAvoidencryption processing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent employs disposable random matrices and one-time noise values that are generated quickly and discarded after a single use. This approach eliminates the need for complex, time-consuming homomorphic encryption operations while maintaining privacy, as the random matrices and noise provide sufficient obfuscation for a single encryption event.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Data Source

PatentUS20240256899A1Vertical federated learning method, apparatus, system and device, and storage medium

Publication Date: 2024.08.01 DOUYIN VISION CO LTD

AI summary

The present disclosure provides a vertical federated learning method, apparatus, system, and device, and a storage medium. The method includes: calculating a noise matrix by a first data party based on a mask matrix, determining a product of a residual vector and the noise matrix as a noise-added residual vector, and sending the noise-added residual vector to a second data party; calculating a gradient vector by the second data party based on the noise-added residual vector to update a model parameter.