Vertical Federated Learning via Encrypted Location Indices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vertical federated learning faces challenges in building and training machine learning models using datasets with different data schemas without direct access to the raw data, as it is difficult to identify corresponding data records across datasets while preserving data privacy.
Innovation Solution
A system and method where a central authority applies transformations to encrypted data from satellite systems to generate location indices, allowing the identification of matching data records across datasets without accessing the raw data, enabling the training of local machine learning models and aggregation into a global model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If encrypted data is transformed to identify matching records across datasets, then data privacy is preserved, but the ability to identify corresponding data records deteriorates
Solution Approach 1:
The patent introduces location indices as an intermediary mechanism that bridges the gap between encrypted data and record identification. The location indices are generated by transforming encrypted data to reveal positional information without exposing the actual data values. This intermediary structure enables the system to identify matching records across datasets while the underlying data remains encrypted and private, thus resolving the contradiction between privacy preservation and record identification capability
Solution Approach 2:
The patent applies parameter changes by transforming encrypted data into a different representation that reveals location information while maintaining data confidentiality. The transformation process changes the state of the encrypted data from a form that protects all information to a form that selectively reveals positional parameters without exposing value parameters. This selective parameter revelation enables record matching while preserving data privacy
2Adaptability or versatility
If datasets with different data schemas are combined for vertical federated learning, then data utility and analytics capability are improved, but the complexity of building and training models increases
Solution Approach 1:
The patent segments the complex task of vertical federated learning into distinct components: (1) generating location indices from encrypted data, (2) using location indices to identify matching records across datasets with different schemas, and (3) training models using the identified matching records. This segmentation breaks down the overall complexity into manageable steps, making the system more tractable while preserving the ability to combine diverse datasets for enhanced analytics
Solution Approach 2:
The location indices serve as an intermediary structure that enables integration of datasets with different schemas without requiring direct access to or understanding of the underlying data values. This intermediary layer abstracts the complexity of schema alignment and record matching, allowing the system to combine diverse data sources while maintaining manageable system complexity
3Reliability
If direct access to raw data is prevented to preserve privacy, then data security is improved, but the ability to build and train machine learning models deteriorates
Solution Approach 1:
The patent introduces location indices as an intermediary that enables model training without direct access to raw data. The location indices provide sufficient information to identify matching records and train models, while the actual raw data remains encrypted and inaccessible. This intermediary mechanism resolves the contradiction by providing just enough information for model training while maintaining data security
Solution Approach 2:
The patent creates a copy of the essential information needed for model training in the form of location indices, without copying or exposing the actual raw data. These location indices are derived from the encrypted data and contain positional information sufficient for training purposes. This copying approach enables model training capability while preserving data security, as the copy contains only minimal necessary information
Data Source
AI summary
Provided herein are systems and methods for vertical federated machine learning. Vertical federated machine learning can be performed by a central system communicatively coupled to a plurality of satellite systems. The central system can receive encrypted data from the satellite systems and apply a transformation that transforms the encrypted data into transformed data. The central system can identify matching values in the transformed data and generate a set of location indices that indicate one or more matching values in the transformed data. The central system can transmit instructions to the satellite systems to access data stored at locations indicated by the location indices and to train a machine learning model using data associated with said locations.