Homomorphic Encryption for Practical Cloud Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques for cloud computing require large computational resources and face challenges in maintaining data confidentiality and privacy, as existing homomorphic encryption methods significantly slow down computations to impractical levels.
Innovation Solution
A method utilizing additively and multiplicatively homomorphic encryption allows partial data encryption, enabling secure and efficient machine learning computations by performing most calculations on encrypted data using a cloud computing system, with a client device handling decryption and encryption while the server performs computational tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is used to process encrypted data in the cloud, then data confidentiality is preserved, but computation speed becomes impractically slow
Solution Approach 1:
The patent segments the data into three distinct sets: training data, field data, and result data. By selectively encrypting only two of these three data sets rather than all data, the system maintains data confidentiality for sensitive information while enabling efficient plaintext processing for non-sensitive data, thus resolving the contradiction between security and computation speed
Solution Approach 2:
The patent applies different encryption states to different parts of the data processing pipeline. Specifically, training data and result data are encrypted while field data remains in plaintext form during processing. This local differentiation allows the system to protect confidential data while maintaining practical computation speeds for operations on non-sensitive data
2Productivity
If data is decrypted before processing in the cloud, then computation speed improves, but data confidentiality is compromised
Solution Approach 1:
The patent extracts the confidentiality requirement from the entire data processing pipeline and applies it selectively only to specific data sets (training data and result data) rather than all data. This extraction allows field data to be processed in plaintext form for improved computation speed while maintaining security for the extracted confidential data sets
Solution Approach 2:
Instead of applying full encryption to all data sets, the patent applies encryption partially - specifically to two out of three data sets (training data and result data). This partial encryption approach provides sufficient security for confidential information while avoiding the performance penalty of encrypting all data, thus resolving the contradiction between speed and security
Data Source
Figure 1
AI summary
A method for evaluating data (28) is based on a computational model, the computational model comprising model data (26), a training function and a prediction function. The method comprises training the computational model by: receiving training data (22) and training result data (24) for training the computational model, and computing the model data (26) from the training data (22) and the training result data (24) with the training function. The method comprises predicting result data (30) by: receiving field data (28) for predicting result data (30); and computing the result data (30) from the field data (28) and the model data (26) with the prediction function. The training data (22) may be plaintext and the training result data (24) may be encrypted with a homomorphic encryption algorithm, wherein the model data (26) may be computed in encrypted form from the training data (22) and the encrypted training result data (24) with the training function. The field data (28) may be plaintext, wherein the result data (30) may be computed in encrypted form from the field data (28) and the encrypted model data (26) with the prediction function.