Machine Learning Model Transfer for Mobile Network Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of deploying and updating machine learning models within wireless mobile communication systems, particularly from the network side to the terminal side, is complex due to the system's composition and numerous manufacturers, requiring a unified solution.
Innovation Solution
The method and apparatus utilize native services to deploy and update machine learning models by establishing AI-native service management components on both the base station and terminal sides, leveraging the user plane and control plane to manage model transfer and installation through a machine learning model library and service registration processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are deployed and updated within wireless mobile communication systems, then network performance and intelligence are improved, but system complexity and deployment difficulty increase due to multiple manufacturers and system composition
Solution Approach 1:
The patent establishes a universal model transfer management mechanism that works across different wireless communication systems (4G, 5G, 6G) and different manufacturers' equipment. The management component provides standardized interfaces and procedures for model transfer, making the deployment process universally applicable throughout the wireless communication ecosystem, thereby resolving the complexity issue while maintaining adaptability.
Solution Approach 2:
The patent introduces a model transfer management component as an intermediary between the model source and the target system. This intermediary handles the complexity of model transfer, validation, and deployment, shielding the underlying system complexity from users while enabling flexible model deployment across different wireless communication systems.
2Speed
If machine learning model parameters are transferred from network side to terminal side, then inference speed and responsiveness are improved, but transmission overhead and air interface burden increase
Solution Approach 1:
The patent extracts the machine learning inference process from the network side and places it at the terminal side. By taking out the inference function from the network infrastructure and embedding it in user equipment, the system achieves faster local inference while reducing the need for continuous network communication during inference operations.
Solution Approach 2:
The patent performs preliminary action by transferring model parameters to the terminal in advance before inference is needed. The model is pre-deployed to the user equipment through the air interface, so when inference is required, the terminal can immediately perform local inference without waiting for model retrieval, thus improving speed while the transmission overhead occurs only during the initial model transfer.
3Adaptability or versatility
If deep neural networks with multiple hidden layers are used to solve complex nonlinear problems, then problem-solving capability is improved, but model size and transmission requirements increase
Solution Approach 1:
The patent enables user equipment to autonomously perform inference using locally deployed deep neural networks. The terminal equipment independently executes the inference process without requiring network assistance during operation, allowing complex multi-layer networks to be deployed locally where they can leverage the device's own computational resources to handle complex nonlinear problems.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A method and apparatus for transferring a machine learning model parameter in a mobile communication system, for achieving deployment/update of a machine learning model when a machine learning deductive model is inside a wireless mobile communication system. The method for transferring the machine learning model parameter in the mobile communication system comprises: an intelligent endogenous service slave management unit in a first device receives a machine learning model file sent by an intelligent endogenous service master management unit in a second device and target functional unit information (S101); and the intelligent endogenous service slave management unit distributes the machine learning model file into a target functional unit in the first device according to the target functional unit information (S102).