Shuffled Secure Multiparty Deep Learning via Data Part Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Secure Multi-Party Computation (SMPC) techniques require high precision circuits to operate on ciphertexts generated using Homomorphic Encryption, which are not available in typical Deep Learning Accelerators (DLAs), preventing their use in secure outsourcing of deep learning computations.
Innovation Solution
The solution involves shuffling randomized data parts across multiple samples, allowing external parties to perform computations without reconstructing the original data, thus eliminating the need for long encryption keys and high precision circuits, enabling DLAs without such circuits to participate in outsourced deep learning computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Homomorphic Encryption is used to protect data privacy in outsourced deep learning computations, then data privacy is improved, but the requirement for high precision circuits and long encryption keys increases device complexity and manufacturing difficulty
Solution Approach 1:
The patent segments deep learning computations into multiple independent parts that can be distributed and executed separately on different devices. By dividing the computation workflow into discrete segments, the system enables secure outsourcing without requiring each individual device to possess high precision circuits, thus resolving the contradiction between data privacy protection and device complexity requirements
Solution Approach 2:
The patent introduces trusted execution environments and secure computation protocols as intermediary layers between data owners and external computing devices. These intermediaries enable privacy-preserving computations by mediating the interaction, allowing standard DLAs without high precision circuits to participate in secure multiparty computations, thereby reducing device complexity while maintaining data privacy
2Reliability
If Homomorphic Encryption with long encryption keys is employed to ensure secure computation, then data privacy is improved, but the computational overhead and processing time increase
Solution Approach 1:
By segmenting computations into smaller independent tasks, the system can process data in parallel batches, reducing the overall computational overhead and processing time while maintaining security through distributed execution, thus resolving the time loss contradiction
Solution Approach 2:
The patent employs preliminary data preprocessing and transformation steps that prepare data in formats optimized for efficient secure computation. By performing preliminary actions such as data encoding and computation graph optimization before outsourcing, the system reduces computational overhead during the actual secure execution phase, thereby decreasing total processing time while maintaining privacy
3Reliability
If high precision circuits are implemented to support Homomorphic Encryption operations, then data privacy protection capability is improved, but manufacturing precision requirements and device cost increase
Solution Approach 1:
The patent divides secure computation into multiple low-precision segments that can be executed on standard DLAs, eliminating the need for high precision circuits. By segmenting the computational workflow and using software-based security mechanisms, the system achieves data privacy protection without increasing manufacturing precision requirements
Solution Approach 2:
The patent uses software simulations and emulations of secure computation protocols that replicate the functionality of high precision Homomorphic Encryption operations without requiring actual high precision hardware circuits. This copying approach enables privacy protection through software layers, avoiding the need for expensive high precision manufacturing
Data Source
AI summary
Protection of access to data samples in outsourcing deep learning computations via shuffling parts. For example, each data sample can be configured as the sum of a plurality of randomized parts. Parts from different data samples are shuffled to mix parts from different samples. One or more external entities can be provided with shuffled and randomized parts to generate results of applying a deep learning computation to the parts. The deep learning computation is configured to allow change of the order between applying the summation and applying the deep learning computation. Thus, results of the external entities applying the deep learning computation to their received parts can be shuffled back for the respective data samples for summation. The summation provides the result of applying the deep learning computation to a respective data sample.