Network Data Analytics Model Training via Parameter Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data isolation issues in communication networks hinder the implementation of accurate data collection, analysis, and training due to privacy and security concerns, preventing the NWDAF network element from exchanging original data with other network elements.
Innovation Solution
A network data analytics network element trains a target model using output results from multiple network elements, reflecting the impact of local models, thereby avoiding data leakage and enabling comprehensive and accurate data collection, analysis, and training without exchanging original data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If original data is exchanged between network elements for centralized data collection and training, then data analysis accuracy is improved, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the necessary model parameters and training results from the centralized training process, allowing these to be distributed to local network elements. This enables accurate data analysis at the local level without requiring the exchange or centralization of sensitive original data, thus maintaining data privacy while achieving analysis accuracy.
Solution Approach 2:
The patent introduces a centralized network data analytics function as an intermediary that performs model training and parameter optimization. This intermediary processes data locally at each network element, aggregates only the necessary model parameters (not raw data), and distributes updated parameters back. This mediator approach enables centralized intelligence while preserving data isolation and privacy.
2Productivity
If original data is exchanged between network elements for centralized data collection and training, then model training comprehensiveness is improved, but data security is compromised
Solution Approach 1:
The patent segments the data processing function into two parts: local data processing at each network element and centralized model parameter aggregation. Each network element processes its local data independently to generate model parameters, which are then aggregated centrally. This segmentation allows comprehensive model training across multiple data sources while maintaining data security through isolation.
Solution Approach 2:
The patent uses model parameters and training results as copies that can be exchanged between the centralized analytics function and local network elements. Instead of exchanging original sensitive data, the system exchanges replicated model parameters that capture the essential information needed for training while preserving the security and isolation of the original data sources.
3Reliability
If data isolation is maintained between network elements, then data privacy is protected, but centralized data analysis and training cannot be implemented
Solution Approach 1:
The centralized network data analytics function serves as an intermediary that bridges data isolation and centralized analysis. It receives model parameters from isolated local elements, performs centralized model training and optimization, and distributes updated parameters back. This intermediary enables centralized analysis capability while respecting and maintaining the data isolation boundaries between network elements.
Solution Approach 2:
The patent transforms the data exchange paradigm by changing from exchanging raw data to exchanging model parameters. This parameter transformation allows the centralized analytics function to perform comprehensive model training across multiple isolated data sources by aggregating parameter information rather than raw data, thus enabling centralized analysis while maintaining data privacy and isolation.
Data Source
AI summary
A data processing method and an apparatus relate to the communication field. The data processing method includes: A first network data analytics network element determines identification information of one or more submodels related to a target model. The first network data analytics network element sends the identification information of the submodel to a network element corresponding to the identification information of the submodel. The first network data analytics network element receives, from the network element, a value that reflects impact of the submodel on the target model, and performs training, to obtain the target model. The data processing method and apparatus provide functions such as accurate data collection, data analysis, and data training.