Privacy Preserving Inference Using CSPRNG Matrix Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for Privacy Preserving Inference in machine learning/neural networks face challenges in maintaining user privacy at the client device while keeping operational parameters private on the server, with homomorphic encryption being computationally expensive and straining hardware and software resources.
Innovation Solution
A method and system for Privacy Preserving Inference using cryptographically secure pseudorandom generators to generate and manipulate matrices, allowing for inference computations without revealing sensitive information from the client device and operational parameters from the server, without relying on public-key encryption, by transmitting seeds and matrices between the client and server.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is used to preserve privacy during inference, then user privacy is protected, but computational cost and hardware/software resource strain increase significantly
Solution Approach 1:
The patent extracts the heavy encryption operations from the inference process by using pre-generated trapdoor functions. The client device performs lightweight operations on encrypted data without needing to execute complex decryption or homomorphic encryption operations during inference, thereby reducing computational cost while maintaining privacy protection.
Solution Approach 2:
The system performs preliminary setup operations to generate trapdoor functions and cryptographic keys before the actual inference process. This preliminary action includes generating public-private key pairs and trapdoor functions that enable efficient private-preserving inference without requiring heavy computational resources during the actual inference execution.
2Reliability
If homomorphic encryption is used to preserve privacy, then sensitive information remains confidential, but system complexity and resource requirements increase
Solution Approach 1:
The patent introduces trapdoor functions as an intermediary mechanism between the client and server. These trapdoor functions act as a mediator that enables the server to perform inference on encrypted data without needing full homomorphic encryption capabilities, thereby reducing system complexity while maintaining data confidentiality through the intermediary cryptographic primitive.
3Reliability
If the server performs inference on encrypted data using homomorphic encryption, then privacy is preserved, but inference accuracy and computational efficiency decrease
Solution Approach 1:
The system creates copies of the trapdoor functions and distributes them to client devices. This allows clients to perform local preprocessing and postprocessing operations on encrypted data without requiring the server to execute heavy homomorphic encryption operations, thereby improving inference efficiency while maintaining privacy preservation through the copied cryptographic primitives.
Data Source
AI summary
The disclosed systems, and methods are directed to a method for Privacy Preserving Inference (PPI) comprising receiving a first set of matrix information from a client device, generating kc−1 matrices by operating a first CSPRNG associated with the server with kc−1 seeds, computing inferences from the set of kc matrices, generating a matrix Ss, generating ks−1 random matrices, computing a matrix Yk<sub2>s </sub2>in accordance with the inference matrix Y, the matrix Ss and the ks−1 random matrices, transmitting a second set of matrix information to the client device, the second set of matrix information includes ks−1 seeds corresponding to the ks−1 random matrices and the matrix Yk<sub2>s</sub2>, receiving a matrix U from the client device, and computing an inference value y from the matrix U.