Cross-Domain VFL Node Selection for Privacy-Preserving AI Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 5G network architectures lack support for cross-domain vertical federated learning (VFL) that ensures data privacy and security during multi-node data sharing for enhanced AI model training.
Innovation Solution
A communication method and apparatus that enable a first node to receive information about second nodes capable of executing a VFL task, determining participating nodes, and facilitating secure data sharing and model training through homomorphic encryption and collaborative gradient calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multi-node data sharing is implemented for AI model training, then model performance is improved, but data privacy and security are compromised
Solution Approach 1:
The patent introduces a secret sharing mechanism where data is divided into multiple shares distributed across different nodes. An intermediary computational process (multiparty computation) allows joint AI model training on these shared secrets without any single node having access to the complete original data, thus enabling productivity improvement while maintaining reliability through cryptographic protection
Solution Approach 2:
The patent segments the original data into multiple confidential shares that are distributed to different nodes. Each node holds only a portion of the data (a share), and the complete information can only be reconstructed through collaborative computation. This segmentation ensures that no single node has access to sensitive information while still enabling joint model training
2Productivity
If cross-domain data sharing is enabled, then data silos are broken and model accuracy is enhanced, but system complexity increases
Solution Approach 1:
The patent develops a universal secret sharing and multiparty computation framework that can be applied across different domains and node types. The same cryptographic protocols and computational mechanisms work regardless of the specific domain (telecommunications, healthcare, finance, etc.), reducing the need for domain-specific customizations and managing system complexity while enabling cross-domain collaboration for improved model accuracy
3Reliability
If local data is utilized at multiple nodes, then data security is maintained, but data silos persist and model performance is limited
Solution Approach 1:
The patent uses multiparty computation as an intermediary mechanism that allows nodes to collectively utilize their local data for model training without transferring or exposing the actual data. The computation is performed on encrypted/shared representations, maintaining data security while achieving the performance benefits of multi-node collaboration through the intermediary computational layer
Data Source
AI summary
A communication method and communication apparatuses are provided. The method includes: receiving first information from a first network element, the first information being used for indicating at least one second node that satisfies a condition for executing a VFL task; and determining from the at least one second node at least one participant node that participates in the VFL task.