Trusted Execution Environment for Secure Deep Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep neural network training methods face challenges in maintaining the security and privacy of sensitive data owned by different entities, as they often require access to combined datasets, which can violate data privacy guarantees and regulatory compliance.
Innovation Solution
A blockchain-enabled decentralized ecosystem using trusted execution environments decrypts and combines encrypted datasets from multiple entities within a secure environment, allowing for the training of deep neural networks while keeping sensitive data confidential and private, utilizing shared private keys and homomorphic encryption for secure computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural network training uses combined datasets from multiple entities, then the learning capabilities and model accuracy are improved, but the security and privacy of sensitive data are compromised
Solution Approach 1:
The patent introduces a trusted execution environment (TEE) as an intermediary component that enables secure data processing. The TEE acts as a mediator between data owners and the training system, allowing combined dataset processing while maintaining individual data privacy through hardware-enforced isolation and encryption mechanisms.
Solution Approach 2:
The system segments data processing operations within isolated execution environments. Each entity's data remains separated in encrypted form within the TEE, and only the necessary computational operations are performed on encrypted data, preventing direct access to raw sensitive information while enabling collaborative training.
2Reliability
If encrypted datasets are decrypted and combined within a trusted execution environment, then secure data sharing and private training are enabled, but computational complexity and system overhead increase
Solution Approach 1:
The trusted execution environment provides self-service security features where the hardware itself enforces encryption and isolation mechanisms without requiring complex software-based security layers. The TEE automatically manages key storage, data isolation, and secure computation, reducing the need for additional security overhead.
3Object-affected harmful factors
If sensitive data remains encrypted during training, then privacy compliance is maintained, but data processing and model training operations become more difficult
Solution Approach 1:
The system changes the state of data from plaintext to encrypted form within the TEE, and the hardware automatically manages the encryption parameters and keys. This parameter change enables privacy compliance while the TEE's hardware acceleration handles the computational complexity of encrypted operations transparently.
Data Source
AI summary
Training a deep neural network model using a trusted execution environment is provided. A selection of two or more encrypted files owned by different entities within a plurality of encrypted files containing sensitive datasets is made by a user of a client device. The two or more encrypted files owned by the different entities are decrypted within the trusted execution environment to form decrypted sensitive datasets owned by the different entities. The decrypted sensitive datasets owned by the different entities are combined within the trusted execution environment to form combined sensitive data owned by the different entities. The deep neural network model is generated within the trusted execution environment based on the combined sensitive data owned by the different entities. The deep neural network model is trained within the trusted execution environment using the combined sensitive data owned by the different entities.