Graph Neural Network Federated Learning for Spatial-Temporal Data Mining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current federated learning methods fail to fully utilize association features between spatial-temporal data, leading to less satisfying model training effects due to the inability to effectively mine these associations during the training process.
Innovation Solution
A federated learning method based on graph neural networks is implemented, where each member device mines graph nodes and relationships in spatial-temporal data to generate graph-structured data, trains a local graph neural network model, and shares update information to update the model across devices, ensuring the association features are fully utilized and local data security is maintained.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional federated learning methods are used to train models on spatial-temporal data, then data security is maintained, but association features between spatial-temporal data cannot be fully utilized
Solution Approach 1:
The patent segments spatial-temporal data into graph-structured representations where data points are divided into nodes and their relationships are segmented into edges. This segmentation allows the model to capture association features between different spatial-temporal data points while maintaining the federated learning framework's data security principles.
Solution Approach 2:
The patent transforms traditional tabular spatial-temporal data into graph-structured data, adding a structural dimension that explicitly represents associations between data points. This dimensionality change enables the model to utilize association features that were previously inaccessible in standard federated learning approaches.
2Productivity
If graph-structured data processing is implemented to mine association features, then model training effectiveness is improved, but computational complexity increases
Solution Approach 1:
The patent implements graph neural network processing partially within the federated learning framework, performing graph construction and feature extraction locally on member devices while keeping the core model training distributed. This partial implementation captures association features without requiring complete graph processing centralization, thus managing computational complexity.
3Reliability
If local data is used for training without centralization, then data security is ensured, but association features across different data sources cannot be fully exploited
Solution Approach 1:
The patent introduces graph-structured data as an intermediary representation that encodes association features from multiple data sources. This intermediary structure allows the model to learn cross-source associations without requiring direct access to raw data from other sources, thus maintaining data security while exploiting association features.
Data Source
AI summary
Implementations of the present specification provide a federated learning method and apparatus based on a graph neural network, and a federated learning system. In the federated learning method, each first member device performs mining of graph nodes and mining of a relationship among graph nodes on local spatial-temporal data to generate graph-structured data; trains a local graph neural network model by using the graph-structured data, to obtain update amount information; and sends the update amount information to a second member device; and the second member device receives the update amount information sent by each first member device; obtains combined update amount information based on the received update amount information; and separately sends corresponding model update information to each first member device based on the combined update amount information for each first member device to update the local graph neural network model based on the corresponding model update information.