Nonorthogonal Data Encryption for Secure Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training artificial neural networks on proprietary or sensitive data face challenges in maintaining data privacy and security, as they often rely on cloud computing and federated learning, which can lead to data devaluation and security risks.
Innovation Solution
The method involves modifying training and test examples using nonorthogonal transformations, such as padding and perturbation, to create modified datasets that are used to train neural networks, ensuring that each transformation is unique and potentially combining with orthogonal transformations, while maintaining data privacy and security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transmitted to cloud computing resources for training neural networks, then computational capability and training efficiency are improved, but data security and privacy are compromised
Solution Approach 1:
The patent introduces an intermediary transformation process that converts original data into transformed data before cloud processing. The transformation layer acts as a mediator between the original data and the cloud computing resources, allowing efficient cloud-based training while preventing direct access to sensitive original data. The transformed data can be processed by cloud resources without exposing the underlying proprietary information.
Solution Approach 2:
The patent applies parameter changes by transforming the data representation through mathematical transformations (such as orthogonal transformations or other invertible transformations). The data parameters are changed from their original form to a transformed form that maintains computational utility while altering the sensitive characteristics. This allows cloud resources to work with transformed parameters without accessing the original sensitive data parameters.
2Adaptability or versatility
If proprietary data is shared with external cloud repositories for training, then neural network training capability is improved, but data value and privacy are devalued
Solution Approach 1:
The patent creates transformed copies of the original data for cloud processing. Instead of sharing the original proprietary data, the system generates transformed versions that serve as functional copies for training purposes. These copies can be processed externally while the original data remains secure and retains its full value and proprietary status.
3Reliability
If federated learning is used to train neural networks locally, then data privacy is improved, but computational resource requirements and system complexity increase
Solution Approach 1:
The patent extracts the transformation logic from the complex federated learning system and consolidates it into a separate preprocessing component. By taking out the transformation function and making it a distinct module, the system reduces overall complexity while maintaining the privacy benefits of local processing. The transformation can be applied locally without requiring complex distributed coordination.
Data Source
AI summary
Systems and methods are disclosed for non-orthogonal encryption of data such that artificial neural networks are trained directly on the encrypted data and testing / inference is performed directly on encrypted data. For use with artificial neural networks, original input tensors are programmatically subjected to a sequence of encryption steps: a method for padding the feature set with randomly chosen sets of values; a method for perturbation of the data by altering elements of the feature set (the values stored in the dimensions of input vectors); a method that applies a fixed index shuffle of the data elements in the feature set or applies a random orthogonal matrix transformation of the data elements; and a method for applying convolutions (filters) and pooling (downsampling) to the fixed shuffled data such that informational structure is preserved.