Neural Network Model Segmentation for Secure TEE Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training neural network models face challenges in ensuring data security, as users who provide training data do not want to share input data, and conventional schemes using trusted execution environments (TEEs) are difficult for machine learning engineers to implement due to the complexity of learning TEE techniques.
Innovation Solution
Divide the neural network model into two parts, with one part processed in a trusted execution environment and the other in an untrusted environment, and utilize syntax conversion and compilation to facilitate secure data protection through a multi-party computation frame.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional trusted execution environment schemes are used to protect training data, then data security is improved, but implementation complexity and difficulty increase significantly
Solution Approach 1:
The neural network model is divided into two distinct parts: a first part processed in a trusted execution environment (TEE) for secure computation, and a second part processed in an untrusted environment for additional processing. This segmentation allows the system to achieve data security through TEE while avoiding the need to implement complete TEE solutions for the entire model, thereby reducing overall implementation complexity.
Solution Approach 2:
The patent introduces a syntax conversion module as an intermediary between the TEE and untrusted environment. This converter handles the communication and data exchange between the two environments, abstracting away the complexity of direct TEE integration. By using this intermediary layer, machine learning engineers can implement secure data processing without needing to deeply understand TEE technical details.
2Reliability
If the entire neural network model is processed in a trusted execution environment, then data security is improved, but the complexity of learning and implementing TEE techniques increases
Solution Approach 1:
By segmenting the model into two parts processed in different environments, the patent reduces the scope of TEE implementation requirements. Engineers only need to implement TEE for the first part rather than the entire model, making the implementation process more manageable and easier to learn.
Solution Approach 2:
The syntax conversion module serves as a mediator that simplifies the interaction between TEE and untrusted environments. This intermediary handles the complexity of environment communication, allowing engineers to focus on model logic rather than TEE implementation details, thereby improving ease of operation.
3Productivity
If users provide training data directly to model trainers, then training efficiency is improved, but data security and user trust deteriorate
Solution Approach 1:
The patent segments the training process into two environmental zones: a TEE zone where data security is maintained and an untrusted zone where additional processing occurs. This allows the system to process training data securely without requiring users to share their data externally, thus maintaining both training efficiency and data security.
Solution Approach 2:
The system enables users to maintain control over their data by processing it within their own TEE environment. The self-service mechanism allows users to trust the system with their data while still maintaining security, as the TEE ensures data remains isolated and protected throughout the training process.
Data Source
AI summary
Illustrative embodiments relate to a method, a device, and a computer program product for generating a neural network model. The method includes dividing the neural network model into multiple parts, wherein the multiple parts include a first part for processing an input to the neural network model and a second part for receiving an output from the first part. The method further includes converting, based on syntax for a trusted execution environment, a first part of code in source code of the neural network model and corresponding to the first part. The method further includes compiling the converted first part of code and a second part of code in the source code and corresponding to the second part; and arranging the compiled first part of code and the compiled second part of code respectively in the trusted execution environment and an untrusted execution environment for generating the neural network model.