Selective Homomorphic Encryption for Private-Variable Machine Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Homomorphic encryption methods face inefficiencies in terms of time and storage when used for machine learning, particularly in processes that can take several days compared to minutes for general text calculations.

Innovation Solution

A method and apparatus that perform regression analysis by encrypting only private variables using homomorphic encryption, involving operations such as deriving classification values, distribution average values, and performing computations with homomorphic encryption, while separating and processing data with and without encryption to optimize efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If homomorphic encryption is applied to all data in machine learning processes, then data security is improved, but computation time increases significantly from minutes to several days

Engineering Contradiction:

Improvedata securityVSAvoidcomputation time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent divides the dataset into two segments: private variables requiring homomorphic encryption and non-private variables processed in plaintext. This segmentation allows selective application of encryption only where necessary for security, avoiding the time penalty of encrypting all data while maintaining protection for sensitive information.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies different processing qualities to different parts of the data: homomorphic encryption is applied locally only to private variables that require security, while non-private variables are processed using standard plaintext operations. This local quality approach optimizes the balance between security and computation speed.

Inventive Principle:

Principle #3Local quality

2Reliability

If homomorphic encryption is used for all variables, then privacy protection is improved, but storage requirements and computational overhead increase

Engineering Contradiction:

Improveprivacy protectionVSAvoidcomputational overhead

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts and identifies private variables from the dataset that require privacy protection, separating them from non-private variables. By taking out only the necessary private variables for encryption, the system reduces computational overhead and storage requirements compared to encrypting all variables, while still maintaining privacy protection where needed.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12425181B2Using homomorphic encryption with private variables

Publication Date: 2025.09.23 SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION

AI summary

Provided is homomorphic encryption data processing method and apparatus and relates to homomorphic encryption data processing method and apparatus that set weighted values to a segment classification value (a logit value) and a distribution value of a dataset including a homomorphic encryption and uses weighted values to perform computation and learning of data in a state in which the homomorphic encryption is maintained.