Fair Secure Vertical Federated Learning Server
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current federated learning systems face challenges in ensuring model fairness, security, accuracy, efficiency, and general applicability, particularly in vertical federated learning where private data and model parameters need to be protected while maintaining high performance.
Innovation Solution
The proposed system implements a fair and secure vertical federated learning (FSVFL) method that processes local model outputs from multiple data owners to adjust global model parameters, ensuring model fairness through fairness constraints, maintaining data privacy by not transmitting private data or parameters, achieving high accuracy without approximations, and operating efficiently with plain text computations, compatible with various machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is shared to improve model accuracy, then model accuracy is improved, but data privacy is compromised
Solution Approach 1:
The patent segments the training data horizontally across multiple data owners, with each owner retaining their private data locally. Only model parameters (gradients, weights) are shared with the central server for aggregation, eliminating the need to centralize sensitive data while achieving collaborative model training
Solution Approach 2:
The patent introduces encrypted model parameters as an intermediary medium between data owners and the central server. Data owners encrypt their local model parameters before transmission, and the server aggregates these encrypted parameters without accessing the underlying private data, thus mediating the information exchange while preserving privacy
2Reliability
If encryption is used to protect data privacy, then data security is improved, but computational efficiency deteriorates
Solution Approach 1:
The patent extracts the encryption operation from the data processing workflow and applies it only to model parameters rather than raw data. This selective encryption approach maintains security for sensitive information while avoiding the computational overhead of encrypting entire datasets during training operations
Solution Approach 2:
The patent changes the parameter being encrypted from raw data to model parameters (gradients, weights). This parameter transformation allows the system to maintain security for the most sensitive information while performing computations on unencrypted data, thereby improving computational efficiency
3Measurement precision
If model parameters are shared to improve model performance, then model performance is improved, but model fairness deteriorates
Solution Approach 1:
The patent implements partial sharing of model parameters by selectively aggregating only certain parameters (gradients, weights) while excluding others that could reveal sensitive information about data distributions. This partial parameter sharing maintains model performance while preventing fairness violations
Solution Approach 2:
The patent incorporates fairness constraints as feedback mechanisms in the model training process. The central server monitors model parameters for fairness violations and provides feedback to data owners to adjust their local training, ensuring that model performance improvement does not compromise fairness
Data Source
AI summary
Servers, methods and systems are disclosed for fair and secure vertical federated learning. Fair and secure vertical federated learning (FSVFL) systems are disclosed that achieve one or more of the following properties: model fairness, high security, high accuracy, high efficiency, and/or high generality. Private data is retained on local computing systems, which share only their model outputs, and a server or a trusted computing system shares only model gradients and randomly partitioned sets of data sample identifiers with untrusted computing systems. A fairness constraint protects protected classes of data samples against model training resulting in bias on the basis of the protected classes.