UE AI/ML Capability Classification for Model Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There are no existing standards for categorizing user equipment (UE) capabilities related to artificial intelligence and machine-learning (AI/ML) in telecommunication networks, leading to integration and verification issues, and inefficient resource utilization during AI/ML model training and inference.
Innovation Solution
A method for classifying UE capabilities based on parameters such as processor type, memory, battery power, and RF hardware, determining a categorical classification, and transmitting appropriate AI/ML models or training data to optimize model training and inference based on UE capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AI/ML modeling is used to optimize telecommunication network functions, then network service quality is improved, but integration and verification issues arise due to lack of capability definition standards
Solution Approach 1:
The patent transforms the capability definition from traditional radio-frequency parameters to a standardized capability level parameter system (level 1-5). This parameter change simplifies the complexity by abstracting multiple hardware specifications into a single capability level indicator that directly correlates with AI/ML model suitability, resolving the integration and verification issues mentioned in the contradiction.
Solution Approach 2:
The patent segments user equipment capabilities into distinct levels (level 1 through level 5), where each level represents a specific range of processing capabilities. This segmentation allows the network to categorize and manage diverse devices systematically, addressing the integration complexity by creating discrete, manageable capability categories rather than dealing with continuous variations in hardware specifications.
2Productivity
If capability categorization is implemented to optimize AI/ML usage, then resource allocation efficiency is improved, but system complexity increases due to additional classification requirements
Solution Approach 1:
The patent implements preliminary capability assessment by determining the user equipment's capability level during initial network registration or capability exchange, before actual AI/ML model deployment. This preliminary classification allows the network to pre-identify suitable devices for specific model types, improving resource allocation efficiency without adding complexity during the actual model execution phase.
Solution Approach 2:
The capability level parameter serves multiple functions simultaneously: it categorizes device capabilities, determines suitable AI/ML model types, guides resource allocation decisions, and simplifies network management. This multi-functionality reduces overall system complexity by using a single parameter for multiple purposes rather than implementing separate classification systems for each function.
3Measurement precision
If full-scale AI/ML training is performed on high-capability devices, then model accuracy is improved, but energy consumption and processing load increase
Solution Approach 1:
The patent applies local quality by matching different training intensities to specific device capability levels. High-capability devices (level 4-5) receive full-scale training for high accuracy requirements, while medium-capability devices (level 2-3) receive lightweight training, and low-capability devices (level 1) receive only inference operations. This localized approach optimizes the balance between model accuracy and energy consumption for each device segment.
Solution Approach 2:
The patent implements partial action by providing lightweight training (subset of training data, limited epochs) to medium-capability devices instead of full-scale training. This partial training approach achieves acceptable model performance for these devices while significantly reducing energy consumption and processing load compared to full training, resolving the contradiction between accuracy and energy use.
Data Source
AI summary
A method, system, apparatus, and non-transitory computer-readable medium for classifying machine-learning capabilities of a user device in a telecommunication network may be provided. The method may be performed by one or more processors, and may include receiving user device capability information from the user device; based on the user device capability information, determining a classification of machine learning capabilities of the user device; and transmitting, to the user device, data associated with a machine learning model based on the classification of the machine learning capabilities of the user device.