Secure Hypersphere for Federated Learning Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning techniques face challenges in securing and protecting sensitive data when sharing embeddings across client devices, as these embeddings can expose private information, compromising security and privacy, especially in applications like biometric authentication and medical data.
Innovation Solution
The method involves generating a secure hypersphere that encompasses the local hypersphere, allowing for the sharing of information about the secure hypersphere instead of the local centroid, thereby protecting the underlying data used to train the machine learning model, while minimizing intra-class variation and maximizing inter-class variation to maintain model accuracy and privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If embeddings are shared across client devices for federated learning, then model training accuracy is improved, but data privacy and security are compromised
Solution Approach 1:
The patent introduces secure embedding centers as an intermediary representation that mediates between the need for collaborative model training and data privacy protection. Instead of sharing raw embeddings directly, the system shares transformed secure centers that preserve training utility while preventing reconstruction of sensitive client data. This intermediary representation allows federated learning to proceed with improved accuracy while maintaining privacy boundaries.
Solution Approach 2:
The patent transforms the original embedding parameters into secure embedding centers through mathematical transformations (such as adding noise, applying transformations, or computing aggregated statistics). This parameter change converts sensitive embedding vectors into protected representations that maintain the essential information needed for model training while removing or obscuring the sensitive client-specific information that could lead to privacy breaches.
2Object-affected harmful factors
If secure transformations are applied to embeddings, then data privacy is protected, but information loss occurs
Solution Approach 1:
The patent applies partial transformation to embeddings, converting only the necessary components into secure representations while preserving essential information. Rather than completely transforming or obfuscating all embedding data, the system selectively transforms portions that contain sensitive information while maintaining the structural and semantic information needed for effective model training. This partial action approach balances privacy protection with information retention.
3Reliability
If federated learning is implemented across multiple devices, then model robustness is improved, but computational complexity increases
Solution Approach 1:
The patent extracts and isolates the essential training information from complex embedding data, separating the useful signal from the sensitive noise. By extracting only the necessary components (secure embedding centers) and leaving behind the sensitive detailed information, the system simplifies the federated learning process while maintaining model robustness. This extraction approach reduces the complexity of coordination and communication between devices.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for training a machine learning model. The method generally includes receiving, at a local device from a server, information defining a global version of a machine learning model. A local version of the machine learning model and a local center associated with the local version of the machine learning model are generated based on embeddings generated from local data at a client device and the global version of the machine learning model. A secure center different from the local center is generated based, at least in part, on information about secure centers shared by a plurality of other devices participating in a federated learning scheme. Information about the local version of the machine learning model and information about the secure center is transmitted by the local device to the server.