Access Network AI Interaction for Privacy-Preserving Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current lack of a defined interaction mode between AI entities in access networks and user equipment hinders the efficient application of AI technology in radio access networks, limiting their processing capabilities.
Innovation Solution
A method is introduced where AI entities in the access network interact with terminal devices to perform training and inference tasks, using AI models that do not include user data for privacy protection, and utilize deep reinforcement learning to monitor performance indicators and update models based on feedback and reward information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AI technology is applied to radio access network, then processing capability is improved, but interaction mode between AI entity and terminal device is undefined
Solution Approach 1:
The patent segments the AI system into two distinct components: an AI entity deployed in the access network and terminal devices with AI capabilities. This segmentation allows each component to have specialized functions - the AI entity handles model management and coordination while terminal devices perform local inference and data collection, thereby defining a clear interaction mode that improves processing capability without excessive complexity
Solution Approach 2:
The patent introduces a standardized interaction protocol as an intermediary layer between the AI entity and terminal devices. This protocol mediates communication by defining specific message formats, interaction flows, and data exchange standards, which enables efficient coordination while maintaining system modularity and reducing overall complexity
2Measurement precision
If user data is included in AI model training, then model accuracy is improved, but user privacy is compromised
Solution Approach 1:
The patent extracts sensitive user data from the AI training process by implementing a federated learning architecture. In this architecture, training data remains localized at terminal devices and is never transmitted to the AI entity. Instead, only model updates and gradients are exchanged, which contain no direct user information. This extraction of sensitive data while preserving training capabilities achieves both model accuracy and privacy protection
Solution Approach 2:
The patent creates copies of the AI model that are distributed to terminal devices for local training. Each terminal device maintains a local copy of the model parameters and performs training computations locally using its own data. This copying approach allows the system to leverage local data for improving model accuracy while ensuring that original user data never leaves the terminal device, thus protecting user privacy
Data Source
Figure 1~2
Figure 3a
Figure 3b
AI summary
Embodiments of this application disclose an information processing method and a related device. Embodiments of this application provide a first AI entity in an access network, and define a plurality of basic interaction modes between the first AI entity and a terminal device. In an interaction mode, the first AI entity may receive second AI model information sent by the terminal device. The second AI model information does not include user data of the terminal device. The first AI entity may update first AI model information of the first AI entity based on the second AI model information, and then send updated first AI model information to the terminal device, so that the terminal device trains and updates the second AI model information. It can be learned that the first AI entity in the access network applies an AI technology to a radio access network. This helps improve a processing capability of the radio access network.