Private AI/ML Model Lifecycle Management via Network Metadata
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Network devices are unable to manage the lifecycle of private AI/ML models deployed on first devices due to lack of knowledge about these models, leading to inefficiencies in model management and performance optimization.
Innovation Solution
A method and apparatus for model management that allows network devices to receive and transmit lifecycle management control information to first devices, utilizing model associated information such as training dataset, performance, quantization, and coexistence capability to manage private models effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If private AI/ML models are deployed on first devices for improving communication system performance, then model performance and system efficiency are improved, but network device's ability to manage the lifecycle of these models deteriorates due to lack of knowledge about the private models
Solution Approach 1:
The patent introduces model associated information as an intermediary that bridges the gap between private models on first devices and the network device. This information includes model identifiers, capability information, and other metadata that enables the network device to understand and manage private models without requiring access to the model weights themselves, thus maintaining privacy while improving manageability
Solution Approach 2:
The patent segments the model management information into distinct components: model associated information (identifier, capability, performance metrics) separate from the actual model weights. This segmentation allows the network device to manage models based on their metadata while the private models remain confidential on the first devices
2Ease of operation
If model associated information is transmitted to enable network device management, then model lifecycle management capability is improved, but information security and privacy protection may deteriorate
Solution Approach 1:
The patent extracts only the necessary management-related information (model identifiers, capability metadata, performance metrics) from the complete model, transmitting only this extracted subset to the network device. The core model weights and confidential training data remain on the first devices, achieving management capability without compromising privacy
Solution Approach 2:
The patent applies local quality by differentiating between information that needs to be shared (model metadata for management purposes) and information that must remain local (private model weights). The model associated information contains only the specific attributes needed for network device management while leaving sensitive data localized
Data Source
Figure 1~2
Figure 3~6
AI summary
Provided in the present disclosure are a model management method and apparatus, and a device, wherein the method is applied to a first device, and the first device is a terminal or a server. The method comprises: sending model-related information to a network device, wherein the model-related information is used for registration or recognition of a first model; and receiving life cycle management control information sent by the network device, wherein the life cycle management control information is used for management of the first model.