Adaptive Trained Model Transmission for 5G UE Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 5G Core networks struggle to efficiently transfer machine learning models to user equipment (UEs) in real-time due to limited processing capacity and varying network conditions, leading to slowed data transfer and resource utilization, especially for applications like real-time translation and autonomous vehicles.
Innovation Solution
A policy control entity in the cellular network determines the most suitable trained model for download based on UE capacity and network transmission parameters, using a predefined data set to optimize data size and resolution, ensuring efficient data transfer by prioritizing important model features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full-resolution models are transferred to UE, then model accuracy is improved, but data transfer time increases and network resources are consumed
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the model resolution (bit depth) based on network conditions and UE capabilities. The system transforms models from full resolution (e.g., 32-bit or 64-bit floats) to reduced resolution representations, optimizing the balance between accuracy and transfer efficiency. This resolves the contradiction by changing the data parameters adaptively rather than using fixed full-resolution transfers.
Solution Approach 2:
The patent implements partial action by transferring only the essential model components and features needed for the specific application task, rather than transferring complete full-resolution models. The system selectively compresses or transfers only critical model parameters, achieving sufficient accuracy for the application while significantly reducing data transfer volume and time.
2Adaptability or versatility
If larger models are downloaded, then model capability is improved, but UE processing capacity and memory resources are overburdened
Solution Approach 1:
The system changes the parameter of model size and complexity by selecting appropriate model resolutions based on UE capabilities. Instead of uniformly deploying large models, the system adapts model parameters (size, precision, complexity) to match the processing capacity of individual UEs, enabling capable models on powerful devices while using lighter models on resource-constrained devices.
Solution Approach 2:
The patent introduces dynamics by making model selection adaptive and flexible rather than static. The system dynamically adjusts which model version is transferred based on real-time assessment of UE capabilities and network conditions. This dynamic approach allows the system to optimize between model capability and device resource consumption for each specific deployment scenario.
3Speed
If network transmission speed is increased, then data transfer time is reduced, but network infrastructure complexity and cost increase
Solution Approach 1:
The patent extracts the bottleneck from the network infrastructure by addressing speed limitations through data compression and reduction techniques rather than investing in faster physical networks. The system extracts the essential information from full-resolution models and transfers only what is necessary, effectively increasing transfer speed without requiring proportional increases in network infrastructure capability.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method for operating a policy control entity in a cellular network, the method comprising: - determining a quality of service parameter for a data packet session in which one trained model from a plurality of trained models is downloaded to a mobile entity, - determining at least one capacity parameter of the mobile entity, - determining a network transmission parameter of the cellular network, - determining said one trained model from the plurality of different trained models based on a dataset which maps different capacity parameters and transmission capabilities to the plurality of trained models, on the determined capacity parameter, and on the network transmission parameter, - determining routing information indicating where said one trained model is accessible for a transmission to the mobile entity, - transmitting the routing information to a session management entity configured to manage the data packet session in the cellular network.