Homomorphic Encryption System for Neural Network Confidentiality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network models face challenges in processing encrypted data due to high processing loads and limitations in available operations, particularly with additive homomorphic encryption, which restricts the execution of non-integer multiplications and division operations, making it unsuitable for floating-point format decimals.
Innovation Solution
An encryption processing system that generates and processes encrypted data by multiplying plaintext data by a power of a predetermined number, allowing for preprocessing and postprocessing to maintain confidentiality while performing operations equivalent to non-integer multiplications, using homomorphic encryption techniques like Paillier encryption to handle neural network operations efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If additive homomorphic encryption is used to process encrypted data, then data confidentiality is maintained, but non-integer multiplications and division operations cannot be executed
Solution Approach 1:
The patent changes the parameter of encryption method from additive homomorphic encryption to Paillier encryption, which supports non-integer multiplications and division operations while maintaining data confidentiality. This parameter change enables the system to perform a wider range of mathematical operations on encrypted data without exposing the plaintext.
2Adaptability or versatility
If fully homomorphic encryption is used to handle both addition and multiplication operations, then operational versatility is improved, but computation amount becomes enormous
Solution Approach 1:
The patent selects Paillier encryption as the homomorphic encryption method, which provides a balanced parameter configuration supporting both addition and multiplication operations with reduced computational overhead compared to fully homomorphic encryption. This parameter optimization enables neural network operations while keeping computation amount manageable.
3Reliability
If neural network processing is performed on encrypted data, then data leakage is prevented, but processing load increases
Solution Approach 1:
The patent introduces an encryption processing device as an intermediary between the client device and server. This intermediary handles the complex homomorphic encryption and neural network processing operations, offloading the computational burden from the client device while maintaining data security throughout the processing pipeline.
4Productivity
If data is transmitted from client device to server for processing, then processing capability is improved, but data leakage risk increases
Solution Approach 1:
The patent applies preliminary action by encrypting the data using Paillier encryption before transmission to the server. This pre-encryption ensures that the data remains confidential throughout the server processing, eliminating the data leakage risk associated with transmitting plaintext data while still enabling the server to perform necessary computations.
Data Source
AI summary
An encryption processing system includes: an encryption data generation device, an encryption processing device, and a processing result utilization device. A first processor of the encryption data generation device is configured to perform preprocessing by generating encrypted data of homomorphic encryption corresponding to data obtained by multiplying plaintext data as a target by a power of a predetermined number of two or more. A second processor of the encryption processing device is configured to perform acquiring the encrypted data, and executing a processing on the encrypted data in an encrypted state to obtain a processing result in the encrypted state. A third processor of the processing result utilization device is configured to perform acquiring the processing result, and postprocessing by decrypting data of the processing result in the encrypted state and by dividing the decrypted data by the power of the predetermined number of two or more.