Terminal-Edge Cloud Training to Reduce Network Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional mobile communication systems face issues such as network load, data synchronization, aging of dispersed data, and non-realtime processing due to large data concentrations at central cloud servers, leading to performance degradation.
Innovation Solution
A distributed training method is implemented using an edge cloud server to identify and utilize multiple user terminals within its local network coverage for training machine learning models, allowing for localized data processing and aggregation of training results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training is performed at a central cloud server, then data concentration enables centralized model training, but network load increases and processing becomes non-realtime
Solution Approach 1:
The patent segments the centralized training process into distributed training tasks across multiple edge cloud servers. Each edge server independently trains machine learning models using local data, then aggregates results at the central cloud server. This segmentation reduces network load and enables parallel processing, improving both reliability through diverse data sources and productivity through concurrent training operations.
2Reliability
If large quantities of data are concentrated at a central cloud server, then centralized training can be performed, but data synchronization and aging issues occur
Solution Approach 1:
The patent implements local quality by allowing each edge cloud server to utilize locally available data for training. Instead of requiring all data to be centralized, each edge server processes data locally, maintaining data freshness and avoiding synchronization issues. The diverse local data sources improve model training quality while preventing data aging problems associated with centralized data concentration.
3Productivity
If data is dispersed across multiple user terminals, then network load is reduced, but data collection and synchronization become complex
Solution Approach 1:
The patent introduces edge cloud servers as intermediaries between user terminals and the central cloud server. These edge servers collect data from multiple user terminals locally, perform preprocessing and initial training, then transmit aggregated results to the central cloud server. This intermediary layer simplifies data management complexity while maintaining network efficiency by reducing direct communication requirements between all terminals and the central server.
Data Source
AI summary
A distributed training method performed between a terminal and an edge cloud server is disclosed. The distributed training method may include identifying a plurality of user terminals located in a local network coverage of the edge cloud server, providing machine learning models to the plurality of user terminals for training and aggregating results of the training of the machine learning models from the plurality of user terminals when the training of the machine learning models is completed.