Federated Graph Neural Network Model Bases for Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning, training graph neural networks across service parties with heterogeneous graph data architectures is challenging due to privacy concerns, as local data cannot be shared, making unified modeling difficult.
Innovation Solution
The method involves determining model bases with shared reference parameters, where each service party constructs a local model by combining these bases in a predetermined mode, processes local training samples, and sends gradient data to a server for fusion and updating, allowing iterative training while maintaining privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If service parties train graph neural networks locally with their own graph data, then privacy protection is achieved, but model consistency and training effectiveness deteriorate due to heterogeneous graph architectures
Solution Approach 1:
The patent segments the graph neural network model into multiple model bases, where each model basis corresponds to a specific graph architecture type. Service parties with different graph architectures can select and combine relevant model bases to construct their local models, ensuring both privacy protection and model consistency across heterogeneous structures
Solution Approach 2:
The patent creates a universal model basis set that can serve multiple graph architecture types. Each model basis is designed to be universally applicable to different graph structures, allowing service parties to combine these universal bases according to their specific needs while maintaining overall model consistency
2Measurement precision
If service parties share their local graph data for unified modeling, then model accuracy improves, but data privacy security deteriorates
Solution Approach 1:
The patent introduces model bases as an intermediary between service parties' local data and the global model. Instead of directly sharing sensitive graph data, service parties use the shared model bases as a mediator to achieve collaborative training, thereby improving model accuracy while protecting data privacy
Solution Approach 2:
The patent creates copies of model parameters and structures (model bases) that can be shared across service parties without exposing the original sensitive data. These copied model bases serve as surrogates that enable collaborative learning while maintaining data isolation
3Productivity
If service parties use different graph data architectures locally, then local data processing effectiveness is maintained, but federated learning implementation difficulty increases
Solution Approach 1:
The patent allows each service party to maintain local model constructions that are optimized for their specific graph architecture types by selecting appropriate model bases. This local quality approach ensures effective local data processing while the shared model basis framework manages the overall federated learning complexity
Data Source
AI summary
A computer-implemented method, medium, and system are disclosed. One example method includes determining multiple model bases by multiple service parties. A respective local service model is constructed by each service party. Respective local training samples are processed by each service party using the respective local service model to determine respective gradient data corresponding to each model basis. The respective gradient data is sent to a server. In response to determining that the first model basis satisfies a gradient update condition, corresponding gradient data of the first model basis received from each service party are combined to obtain global gradient data corresponding to the first model basis. The global gradient data is sent to each service party. Reference parameters in local model basis corresponding to the first model basis are updated by each service party using the global gradient data to train the respective local service model.