Neural Network Parameter Conversion for Encrypted Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing prediction model conversion methods, such as SecureML, CryptoNets, and MOBIUS, face issues with low prediction accuracy and high computational complexity, making them impractical for secure neural network processing.

Innovation Solution

A prediction model conversion method that converts neural network parameters for homogenization processing into parameters capable of performing nonlinear processing, generating an encrypted model that maintains input secrecy during prediction processing, thereby reducing computational overhead and improving accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing prediction model conversion methods (SecureML, CryptoNets, MOBIUS) are used to perform prediction processing while keeping data secret, then privacy protection is achieved, but prediction accuracy significantly decreases and computational complexity increases

Engineering Contradiction:
Improveprivacy protectionVSAvoidprediction accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent converts multiple homogenization parameters into a single parameter that performs both homogenization and nonlinear processing. This parameter transformation enables the encrypted prediction model to maintain higher prediction accuracy by reducing the complexity of operations on encrypted data, while still preserving privacy through encryption.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If existing prediction model conversion methods (SecureML, CryptoNets, MOBIUS) are used to perform prediction processing while keeping data secret, then privacy protection is achieved, but computational complexity increases

Engineering Contradiction:
Improveprivacy protectionVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges multiple homogenization parameters into a single parameter that performs both homogenization and nonlinear processing functions. This consolidation reduces the number of computational operations required in the encrypted prediction model, thereby decreasing computational complexity while maintaining privacy protection through encryption.

Inventive Principle:
Principle #5Merging (Combining)

3Reliability

If multiple parameters for homogenization processing are used in the prediction model, then processing completeness is achieved, but the amount of computation increases

Engineering Contradiction:
Improveprocessing completenessVSAvoidprocessing speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent transforms multiple homogenization parameters into a single parameter that performs both homogenization and nonlinear processing. This parameter change reduces the computational burden and increases processing speed, while the converted parameter maintains the necessary processing completeness for accurate predictions in the encrypted domain.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3910615B1Prediction model conversion method and system
Publication Date: 2024.08.07 PANASONIC INTELLECTUAL PROPERTY CORP OF AMERICA
  • EP3910615B1 patent drawingFigure 1~2
  • EP3910615B1 patent drawingFigure 3~4
  • EP3910615B1 patent drawingFigure 5~6

AI summary

A prediction model conversion method includes: converting a prediction model by converting at least one parameter which is included in the prediction model and is for performing homogenization processing into at least one parameter for performing processing including nonlinear processing, the prediction model being a neural network (S001); and generating an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted (S002).