Cloud-Edge Model Processing with Local Gradient Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cloud service systems face inefficiencies in updating machine learning models due to high computing demands on cloud servers and edge devices, leading to reduced operating efficiency and data security issues.
Innovation Solution
Implementing a local server between the cloud server and edge devices to collaboratively update models using gradient values, reducing reliance on cloud server and edge device computing capabilities and minimizing data exchange.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the cloud server updates the machine learning model based on computing data from all terminal devices, then the model precision is improved, but the computing amount of the cloud server is increased and operating efficiency is reduced
Solution Approach 1:
The patent segments the model update process by introducing local servers that aggregate gradient values from multiple terminal devices. Instead of the cloud server directly collecting data from all terminals, local servers perform preliminary aggregation, dividing the computing task into distributed segments that reduce the cloud server's burden while maintaining model precision through collaborative federated learning.
2Productivity
If terminal devices perform federated learning to update the machine learning model, then the cloud server's computing amount is reduced, but the computing amount of the terminal device is increased
Solution Approach 1:
The patent introduces local servers as intermediary components between terminal devices and the cloud server. These local servers aggregate gradient values from multiple terminals and perform preliminary model updates, reducing the direct computing burden on terminal devices while maintaining the benefits of federated learning by keeping data locally and minimizing cloud server involvement in heavy computations.
3Measurement precision
If latest computing data is sent to the cloud server for model updates, then the model precision is improved, but data security risks are increased
Solution Approach 1:
The patent extracts and processes only the essential gradient values needed for model updates while leaving the actual computing data localized at terminal devices and local servers. This extraction approach maintains model precision by transmitting the necessary update information to the cloud server while eliminating the need to transfer sensitive raw data, thereby reducing data security risks.
Data Source
AI summary
In a model processing method, a first local server which is disposed between a cloud server and an edge device obtains a data set of the edge device. The data set comprises data used when the edge device performs computing by using a first model provided by the cloud server. The first local server determines, based on the data set of the edge device, a first gradient value for updating the first model, and sends the first gradient value to the cloud server for use by the cloud server to update the first model.