Trusted Execution Environment for Secure Data Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning solutions are vulnerable to the quality of training data, which can be compromised by malicious parties providing incorrect data, leading to suboptimal training and potentially harmful outcomes, such as incorrect classification decisions in self-driving cars.
Innovation Solution
An apparatus and method utilizing a trusted execution environment (TEE) to determine characteristics of data items, obtain cryptographic signatures or hash values, and provide them to nodes for verification, ensuring that only data with desired characteristics is used in computation processes, thereby enhancing the quality and dependability of machine learning outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is obtained from diverse sources to improve machine learning quality, then the quality and diversity of training data is improved, but the risk of data compromise by malicious parties increases
Solution Approach 1:
The patent applies preliminary action by verifying data characteristics and obtaining cryptographic signatures before the training data is used in machine learning processes. The TEE determines characteristics of data items and obtains cryptographic signatures in advance, ensuring data integrity before it enters the training pipeline, thus preventing malicious data from compromising the model.
Solution Approach 2:
The patent introduces a trusted execution environment (TEE) as an intermediary between diverse data sources and the machine learning training process. The TEE acts as a mediator that verifies data characteristics and provides cryptographic authentication, isolating the ML system from direct exposure to potentially compromised data sources while maintaining data quality requirements.
2Reliability
If cryptographic verification processes are implemented to ensure data integrity, then data authenticity is improved, but the computational overhead and processing time increase
Solution Approach 1:
The patent extracts the computationally intensive cryptographic verification operations into a separate trusted execution environment (TEE). By moving data characteristic determination and signature verification into the TEE, the main machine learning training process can proceed with verified data without bearing the full computational burden, thus balancing authenticity verification with processing throughput.
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
According to an example aspect of the present invention, there is provided an apparatus configured to determine, using a first executable program and in a trusted execution environment of the apparatus, at least one characteristic of a data item, obtain, in the trusted execution environment, responsive to a result of the determining of the at least one characteristic, a cryptographic signature over the data item or a hash value obtained from the data item and a hash value obtained from the first executable program, the cryptographic signature being obtained using a key of the trusted execution environment, and provide at least one of the data item or the cryptographic signature to a node.