Confidential MSB Normalization for Accurate Secret-Shared Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In secure computation, shifting the most significant bit (MSB) to align it with a predetermined position results in loss of small inputs and reduced accuracy due to the product sum's high bit output.
Innovation Solution
A secure MSB normalization system and method that includes distributed processing apparatuses with bit decomposition, logical sum acquisition, shift amount acquisition, and shift units to shift the MSB of a vector to a predetermined bit position using (k, n)-secret sharing, ensuring accuracy by shifting the entire vector all at once.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If floating-point numbers are used to represent real numbers in neural network computations, then calculation precision is improved, but computational cost and processing time increase significantly
Solution Approach 1:
The patent changes the numerical parameter representation from floating-point to fixed-point format. Specifically, it uses Q-format fixed-point numbers (e.g., Q1.31 for normalization factors, Q0.31 for correlation values) to represent real numbers, transforming the computational paradigm while maintaining adequate precision for signal processing tasks.
Solution Approach 2:
The patent applies different fixed-point precision levels to different computational components based on their specific requirements. For example, normalization factors use Q1.31 format while correlation values use Q0.31 format, optimizing the balance between precision and computational efficiency for each specific operation.
2Reliability
If confidential computing techniques are implemented to protect intellectual property, then security is improved, but system complexity increases
Solution Approach 1:
The patent introduces a trusted execution environment (TEE) as an intermediary layer between the neural network model and the processing device. The TEE acts as a secure container that holds the intellectual property (normalization parameters and model weights) and performs computations internally, preventing external access while enabling secure collaboration.
Solution Approach 2:
The patent segments the computing system into distinct functional components: a confidential computation service running within the TEE that handles sensitive operations, and external processing devices that perform non-sensitive computations. This segmentation isolates intellectual property protection to only where needed.
3Reliability
If data is processed in encrypted or confidential form, then data security is improved, but processing capability and accuracy deteriorate
Solution Approach 1:
The patent extracts sensitive data (normalization parameters and model weights) into the confidential TEE environment for storage and computation, while keeping the processing device working with less sensitive intermediate results. This extraction allows high-precision arithmetic to occur in the secure environment without compromising overall system security.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A secure MSB normalization system includes n distributed processing apparatuses, in which each of the n distributed processing apparatuses includes a bit decomposition unit, a logical sum acquisition unit, a shift amount acquisition unit, and a shift unit, the n bit decomposition units decompose a vector [[→a]]P of a (k, n)-secret shared share into bits and obtain a bit representation [[→a]]2^L of the vector [[→a]]P, the n logical sum acquisition units obtain a logical sum [[Ai]]2 of all elements for a vector [[→ai]] at each bit position of the bit representation [[→a]]2^L, the n shift amount acquisition units obtain a share <<ρ>>p obtained by distributing a shift amount p for shifting the most significant bit of a logical sum [[A0]]2, ..., [[AL-1]]2 to a fixed position by (k, n)-replica secret sharing by a modulus p, and the n shift units obtain a vector [[2ρ→a]]p in which each element of the vector [[→a]]p is shifted left by p bits.