Private DNN Inference Using Secure Comparison Protocols
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing secure inference protocols for deep neural networks (DNNs) face challenges in efficiently computing non-linear layers like ReLU and Maxpool, and accurately performing fixed-point arithmetic, leading to significant accuracy losses and computational inefficiencies.
Innovation Solution
The development of new semi-honest secure 2-party cryptographic protocols that enable efficient secure computation of DNN inference, specifically by introducing a new comparison protocol for non-linear layers and protocols for division, which are essential for linear layers like convolutions and average pooling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing secure inference protocols are used for DNNs, then privacy preservation is achieved, but computational efficiency and accuracy for non-linear layers deteriorate
Solution Approach 1:
The patent segments the DNN computation into separate linear and non-linear layer protocols. Linear layers use efficient homomorphic encryption operations, while non-linear layers use comparison protocols with oblivious transfer. This segmentation allows each layer type to be optimized independently, resolving the contradiction between privacy preservation and computational efficiency.
Solution Approach 2:
The patent changes the computational parameters by using homomorphic encryption for linear algebra operations and comparison protocols with oblivious transfer for non-linear operations. This parameter change enables efficient secure computation by matching the cryptographic primitive to the operation type, improving both efficiency and accuracy.
2Reliability
If existing secure inference protocols are used for DNNs, then privacy preservation is achieved, but accuracy for fixed-point arithmetic deteriorates
Solution Approach 1:
The patent introduces comparison protocols as intermediaries to handle fixed-point arithmetic accurately. These protocols use oblivious transfer to securely compute comparisons between fixed-point numbers, ensuring accurate arithmetic operations while maintaining privacy. The intermediary comparison protocol resolves the accuracy issue without compromising privacy.
3Measurement precision
If secure computation is performed with detailed comparison protocols, then accuracy is improved, but communication complexity increases
Solution Approach 1:
The patent applies partial action by using comparison protocols only where necessary (for non-linear layers and fixed-point arithmetic) rather than applying full secure computation to all operations. This selective approach maintains accuracy for critical operations while reducing overall communication complexity for the entire DNN inference process.
Data Source
AI summary
A secure inference over Deep Neural Networks (DNNs) using secure two-party computation to perform privacy-preserving machine learning. The secure inference uses a particular type of comparison that can be used as a building block for various layers in the DNN including, for example, ReLU activations and divisions. The comparison securely computes a Boolean share of a bit representing whether input value x is less than input value y, where x is held by a user of the DNN, and where y is held by a provider of the DNN. Each party computing system parses their input into leaf strings of multiple bits. This is much more efficient than if the leaf strings were individual bits. Accordingly, the secure inference described herein is more readily adapted for using in complex DNNs.