Homomorphic Image Inference Using CinS Encoding and Bootstrapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing homomorphic encryption technologies face significant overhead when performing computations on image-type data, leading to potential data leaks and inefficiencies in processing.
Innovation Solution
A server device equipped with a communication unit, memory, and processor that performs CinS encoding on images, converts data using AI models for convolutional and activation function computations, and employs bootstrapping to efficiently process homomorphic ciphertext, reducing computational burden and ensuring privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption technology is used to perform computation on encrypted image data, then data security and privacy are improved, but computational overhead and processing time increase significantly
Solution Approach 1:
The patent segments the computational process into distinct phases: encoding phase (converting images to CinS-encoded data before encryption), encryption phase (applying homomorphic encryption), and computation phase (performing AI model operations on ciphertext). This segmentation allows optimization of each phase independently, reducing overall computational overhead while maintaining security.
Solution Approach 2:
The patent applies CinS encoding to image data before encryption, as a preliminary action. This pre-processing step transforms the data into a format that enables more efficient homomorphic computation, reducing the computational burden during the encrypted processing phase while ensuring data security throughout.
2Reliability
If standard homomorphic encryption is applied to image data, then privacy protection is improved, but data processing efficiency deteriorates
Solution Approach 1:
The patent changes the data representation parameters by introducing CinS encoding (Coefficient-encoded to Slot-encoded transformation). This parameter change optimizes the structure of encrypted image data, enabling faster AI model computations while maintaining strong privacy protection through homomorphic encryption.
Solution Approach 2:
The patent introduces CinS-encoded data as an intermediary form between standard image data and encrypted data. This intermediary representation facilitates efficient homomorphic computation by AI models while preserving privacy, acting as a bridge that improves processing efficiency without compromising security.
3Measurement precision
If multiple encoding conversions are performed during AI model computation, then computational accuracy is improved, but operational complexity increases
Solution Approach 1:
The patent merges the CinS encoding conversion operations with the AI model computation layers. Instead of performing separate encoding conversions between layers, the transformations are integrated into the convolutional and fully connected layer operations, reducing the number of distinct operations while maintaining computational accuracy.
Solution Approach 2:
The patent designs the AI model to handle multiple data formats (CinS-encoded and slot-encoded data) within a unified architecture. The model performs both encoding conversions and computations using the same computational graph, reducing operational complexity while maintaining accuracy through multi-functional layer designs.
Data Source
AI summary
Disclosed is a server device. The server device includes a communication unit configured to perform communication with an electronic device; a memory configured to store an artificial intelligence (AI) model for performing computation in an encrypted state; and a processor, in which the processor is configured tobased on receiving homomorphic ciphertext of CinS-encoded data obtained by performing CinS encoding on a plurality of images from the electronic device through the communication unit, transmit the encryption computation result to the electronic device through the communication unit by obtaining an encryption computation result by inputting the homomorphic ciphertext to the AI model, wherein. the AI model performs each of convolutional computation based on the CinS-encoded data and activation function computation based on slot-encoded data converted from the CinS-encoded data a plurality of times.