AI Model Inference on Encrypted Data via Homomorphic Encryption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for performing inference using artificial intelligence (AI) models on external devices face challenges in protecting user data from leakage during transmission and processing, especially when the external device with lower performance requests a higher-performance device to perform inference on encrypted data, which may lead to data exposure risks.
Innovation Solution
An electronic device is configured to transform the AI model to process encrypted data efficiently, generate parameter information for secure encryption, and perform inference on the encrypted data received from an external device, using homomorphic encryption to ensure data security and reduce computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the external device transmits input data to another device for AI inference, then the inference can be performed using a high-performance device, but the input data may be leaked during transmission and processing
Solution Approach 1:
The patent transforms the AI model parameters and structure to operate directly on encrypted data. By changing the model's computational parameters and using homomorphic encryption properties, the system enables inference without decrypting input data, thus maintaining data security while performing complex AI processing on a high-performance device.
Solution Approach 2:
The patent introduces homomorphic encryption as an intermediary mechanism that allows computations to be performed on encrypted data. This intermediary enables the high-performance device to process AI inference requests while the encryption layer protects the input data from leakage during transmission and processing.
2Reliability
If the AI model processes encrypted data directly, then data security is improved, but the computational overhead increases
Solution Approach 1:
The patent optimizes the AI model parameters specifically for encrypted data processing. By transforming the model architecture and computational parameters to work natively with homomorphically encrypted data, the system reduces the computational overhead compared to traditional approaches that would require decryption and re-encryption cycles.
Solution Approach 2:
The patent performs preliminary transformation of the AI model before inference to prepare it for encrypted data processing. This pre-processing of the model structure and parameters enables more efficient computation during the actual inference phase, reducing the energy consumption required for processing encrypted data.
3Reliability
If the AI model is transformed to process encrypted data, then data security is maintained, but the model complexity increases
Solution Approach 1:
The patent transforms the AI model parameters and computational operations to accommodate encrypted data input. By systematically changing the model's mathematical parameters and using homomorphic encryption properties, the transformation maintains the model's functionality while enabling secure processing, without requiring fundamentally complex architectural changes.
Data Source
AI summary
Provided is a method for an electronic device to perform inference based on encrypted data received from an external device, using an artificial intelligence (AI) model, the method including: transforming the AI model to perform inference based on encrypted data, generating parameter information including information about at least one parameter for encrypting data to be input to the AI model, based on the transformed AI model, transmitting the parameter information to the external device, receiving, from the external device, data encrypted based on the parameter information, and obtaining an inference result output from the transformed AI model by inputting the encrypted data to the transformed AI model.