Client-Side Reference Data Splitting for Secure Cloud ML Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cloud service security measures for machine learning processes are inadequate, as they rely on cloud service providers and are vulnerable to human errors leading to potential information leakage.
Innovation Solution
A system where a client apparatus generates reference data and difference data from machine learning data, with the difference data stored on a cloud server, allowing for secure encryption and decryption without relying on the cloud service provider, using a client-side data encryption apparatus and a cloud-side data decryption apparatus.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is encrypted using traditional cloud service provider security measures, then communication and storage security is improved, but the system still depends on the cloud service provider and is vulnerable to human errors
Solution Approach 1:
The patent segments the original data into multiple pieces using secret sharing technology, distributing them across different storage locations. This segmentation ensures that no single point of failure can compromise the entire dataset, and the system can operate independently of any single cloud service provider's security measures
Solution Approach 2:
The patent introduces a secret sharing mechanism as an intermediary layer between the user and the cloud service provider. This intermediary ensures that even if the cloud service provider's security is compromised, the actual data remains protected because the provider only stores encrypted segments without the ability to reconstruct the original data
2Productivity
If all machine learning data is stored in the cloud, then data accessibility and processing capability is improved, but storage security risk increases due to potential information leakage
Solution Approach 1:
The patent divides machine learning data into multiple encrypted segments before storing them in the cloud. This segmentation allows the data to be distributed across multiple storage locations for improved accessibility, while simultaneously reducing the risk of information leakage because each segment alone is useless without the others
Solution Approach 2:
The patent applies encryption and secret sharing techniques before data is uploaded to the cloud, performing a preliminary protective action. This ensures that even if the cloud storage is compromised, the data remains protected because the harmful factor (information leakage) has already been counteracted by the encryption layer
3Reliability
If data is divided into multiple pieces for security, then security against information leakage is improved, but data restoration complexity increases
Solution Approach 1:
The patent implements a feedback mechanism where the system automatically manages the segmentation and reconstruction processes. The secret sharing algorithm provides built-in feedback mechanisms that track which segments are available and automatically coordinate their assembly, reducing the perceived complexity for users while maintaining high security
Data Source
AI summary
A system is a system including: a cloud server configured to perform a machine learning process; and a client apparatus configured to communicate with the cloud server. The client apparatus includes: a generating unit that generates one or a plurality of reference data from a plurality of data used for the machine learning and that generates a plurality of difference data, wherein the reference data is a reference for at least a part of the plurality of data, and each difference data indicates a difference between each of the plurality of data and corresponding reference data out of the one or the plurality of reference data; and a storage unit that stores the plurality of difference data in a storage apparatus of the cloud server.