WTRU AI Measurement Logging for Immediate Model Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current measurement reporting mechanisms in wireless networks do not enable immediate AI/ML predictions due to non-consecutive or past measurements, necessitating separate training and inference operations, which can be inefficient for mobile devices with computation resources.
Innovation Solution
Enhanced measurement and logging procedures that allow wireless transmit/receive units (WTRUs) to transmit AI/ML capability information, receive models, and perform conditional logging and training based on accuracy thresholds and location, enabling immediate model validation and reporting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If legacy measurement procedures are used, then measurement data can be collected, but the data is non-consecutive or contains past measurements which prevents immediate AI/ML predictions
Solution Approach 1:
The patent implements preliminary action by having WTRUs perform AI/ML training operations in advance using collected measurement data while in idle or inactive states. The trained models and parameters are prepared beforehand and stored locally, enabling immediate inference operations when needed without waiting for network coordination or data transmission delays.
Solution Approach 2:
The patent applies self-service by enabling WTRUs to autonomously perform AI/ML training and inference operations using their own computation resources and locally stored measurement data. The devices self-manage the training process, model validation, and parameter updates without requiring continuous network intervention, thereby eliminating time delays associated with centralized processing.
2Reliability
If separate training and inference operations are performed, then AI/ML models can be developed, but the process becomes inefficient for mobile devices with limited computation resources
Solution Approach 1:
The patent merges training and inference operations by enabling WTRUs to perform both functions using the same local computation resources and measurement data. The training phase uses idle/inactive state data collection, while the inference phase uses the same devices for real-time predictions, eliminating the need for separate centralized training infrastructure and improving overall system efficiency.
Solution Approach 2:
The patent segments the AI/ML operations into distinct phases that can be executed independently: data collection during idle/inactive states, training operations when computation resources are available, and inference operations when predictions are needed. This segmentation allows mobile devices to efficiently manage their limited resources by performing different operations at appropriate times rather than requiring all operations simultaneously.
3Productivity
If WTRUs transmit AI/ML capability information and perform conditional logging, then resource utilization is optimized, but the system complexity increases
Solution Approach 1:
The patent changes operational parameters by introducing conditional logging that adapts to AI/ML model requirements. The system dynamically adjusts logging behavior based on parameters such as model accuracy thresholds, data sufficiency conditions, and specific measurement requirements, allowing efficient resource utilization while managing complexity through parameter-based control rather than structural complexity.
Data Source
AI summary
Procedures, methods, architectures, apparatuses, systems, devices, and computer program products directed to artificial intelligence-specific idle/inactive/connected mode measurements procedure. In an embodiment, a method implemented by a wireless transmit receive unit (WTRU), the method comprising: receiving, from a network, a first message comprising a configuration about AI/ML model training and associated measurements and logging periodicity; performing minimization of drive test (MDT) measurements; selecting an AI/ML model for training based on the based on the first message; training the selected AI/ML model based on MDT measurements; and in response to accuracy of the trained model above a configured accuracy threshold, triggering transition to connected state and reporting to the network the trained AI/ML model identity and AI/ML model parameters.