Inference Time Management for Intelligent Models in Wireless Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wireless communication systems face challenges in efficiently managing intelligent (AI/ML) models and functionalities, particularly in determining the inference time required for these models, which varies based on hardware configurations and internal model structures.
Innovation Solution
The proposed solution involves an apparatus and method for intelligent model and functionality management in wireless communication systems, where user equipment and base stations share inference time information to identify and manage supported intelligent models and functionalities. This includes processes for computing minimum required inference time, transmitting this information, and performing life cycle management of intelligent models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If inference time information is not shared between base station and user equipment, then system complexity is reduced, but intelligent model selection and management efficiency deteriorates
Solution Approach 1:
The patent applies preliminary action by having user equipment compute and report inference time information in advance during capability reporting, before actual intelligent model execution. This allows the base station to pre-evaluate and select appropriate models based on reported inference times, avoiding the need for real-time computation and selection during model execution, thus improving selection efficiency without adding significant system complexity
Solution Approach 2:
The patent uses inference time information as an intermediary parameter that mediates between user equipment capabilities and base station model selection decisions. This intermediary metric enables the base station to evaluate and select suitable intelligent models without direct complex interactions with user equipment hardware, simplifying the overall system architecture while maintaining effective model management
2Measurement precision
If comprehensive inference time information is collected from all user equipment, then model performance matching is improved, but information transmission overhead increases
Solution Approach 1:
The patent extracts only the essential inference time information from user equipment capability data, separating this critical parameter from other capability details. By extracting and transmitting only the inference time metric rather than comprehensive capability information, the system achieves accurate model-performance matching while minimizing transmission overhead
Solution Approach 2:
The patent applies local quality by focusing inference time information collection and reporting on specific user equipment that require intelligent model execution, rather than universally collecting detailed capability information from all equipment. This localized approach ensures precise model matching for relevant devices while reducing overall system information overhead
3Reliability
If intelligent models with longer inference time are selected, then model accuracy is improved, but transmission delay increases
Solution Approach 1:
The patent applies dynamics by enabling the base station to dynamically select intelligent models based on real-time service requirements and the reported inference time capabilities of user equipment. This dynamic selection allows the system to choose higher accuracy models when time permits and lower accuracy models when delay constraints exist, optimizing the balance between model accuracy and transmission delay according to specific service conditions
4Adaptability or versatility
If user equipment with limited hardware resources runs complex intelligent models, then model functionality is improved, but system reliability deteriorates due to insufficient performance
Solution Approach 1:
The patent applies parameter changes by using inference time information as a key parameter to evaluate and match user equipment hardware capabilities with appropriate intelligent model complexity. The base station adjusts model selection based on reported inference time parameters, ensuring that models with functionality appropriate to each device's hardware resources are selected, thereby maintaining both adaptability and execution reliability
Data Source
AI summary
Proposed is a wireless communication system and, in more detail, an apparatus and method for intelligent model and functionality management considering inference time in a wireless communication system. A method of operating user equipment (UE) for identifying and managing an intelligent functionalities or models in a wireless communication system, includes a process of receiving UE capability enquiry from a base station (BS), and a process of transmitting UE capability information to the base station, wherein the UE capability information includes inference time information, the base station identifies intelligent functionality and model that the user equipment can support, on the basis of the UE capability enquiry and the UE capability information, and performance of an intelligent model that the user equipment can perform is identified on the basis of the inference time information.