FHE Network Encoding via Scaling and Shift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for converting conventional computational networks into fully homomorphic encryption (FHE) networks face challenges due to value limitations in FHE schemes, leading to accuracy loss and increased computational resources.
Innovation Solution
The method involves encoding values from the first computational network into a range supported by the FHE scheme, using a scaling factor and shift to map values into a prescribed range, allowing operations to be performed directly on encrypted values without the need for programmable bootstrapping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If values from the first computational network are directly used in the FHE scheme, then the computational network can be converted to FHE operations, but accuracy is lost due to value range limitations in FHE schemes
Solution Approach 1:
The patent applies parameter changes by introducing scaling factors and shift values to transform the range of computational values. The encoding function E(x) = (x - offset) * scale maps values from the original range to the FHE-supported range, allowing accurate representation of values that would otherwise exceed FHE limitations while maintaining computational correctness
2Ease of operation
If programmable bootstrapping is used to handle value range limitations, then FHE operations can be performed, but computational overhead and complexity increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-computing scaling factors and shift values during the network conversion phase. These encoding parameters are determined beforehand and embedded into the FHE network structure, eliminating the need for runtime bootstrapping operations and enabling direct homomorphic evaluation with reduced computational overhead
Solution Approach 2:
The patent extracts the value range adaptation problem from the runtime computation phase and relocates it to the network conversion phase. By separating the encoding parameter computation from the homomorphic evaluation, the patent eliminates complex bootstrapping operations during execution and achieves simpler, faster FHE operations
3Manufacturing precision
If large cryptographic parameters are used to maintain accuracy in FHE, then computational accuracy is preserved, but computational resources and overhead increase
Solution Approach 1:
The patent optimizes the balance between accuracy and resources by carefully selecting scaling factors and shift values that map the original value range efficiently into the FHE range. This optimized parameter selection maintains computational accuracy while using minimal cryptographic parameters, reducing the size of FHE ciphertexts and the computational resources required for homomorphic operations
Data Source
Figure 1a~1d
Figure 2
Figure 3
AI summary
Some embodiments are directed to a computer-implemented method to convert a first computation network of operations into a second computation network of operations. The operations of the second computation network being fully homomorphic encrypted (FHE) operations. The converting may include obtaining for the input values of the operation a supported range of the input value in the first computation network, and determining a prescribed range for the input values of a corresponding FHE operation.