Network Model Supervision with Data-Usage Status Signaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network data analytics functions (NWDAF) face challenges in accurately evaluating model performance information due to variations in network operations not based on analysis results, leading to poor accuracy in performance evaluation.
Innovation Solution
A model supervision processing method involving network devices that exchange information to indicate the usage status of data analysis results, allowing for improved evaluation of model performance by filtering out data that affects network operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If network data analytics function uses model for task analysis to obtain analysis result, then task analysis can be completed, but model performance evaluation accuracy deteriorates due to network operations not determined based on analysis results affecting the evaluation data
Solution Approach 1:
The patent extracts and separates the evaluation data collection process from general network operations. It specifically extracts only those network operations that are actually determined based on analysis results, excluding operations that are not influenced by analysis results. This extraction principle resolves the contradiction by ensuring that evaluation data reflects only the relevant operations that demonstrate actual model performance impact.
Solution Approach 2:
Instead of collecting all network operations and filtering out irrelevant ones, the patent inverts the approach by directly identifying and collecting only the operations that are determined based on analysis results. This inversion transforms the problem from a filtering task to a selection task, improving evaluation accuracy by focusing exclusively on operations that truly reflect model performance.
2Quantity of substance
If all network operations are collected for model performance evaluation, then comprehensive data is obtained, but evaluation accuracy deteriorates due to inclusion of operations not affected by data analysis results
Solution Approach 1:
The patent applies the extraction principle by removing irrelevant network operations from the evaluation dataset. It specifically extracts only the subset of operations that are determined based on analysis results, thereby maintaining sufficient data volume for reliable evaluation while eliminating operations that would dilute the evaluation accuracy.
Solution Approach 2:
The patent applies local quality by differentiating between types of network operations and assigning different evaluation weights or inclusion criteria to each type. Operations that are determined based on analysis results are given high priority and included in evaluation, while operations not influenced by analysis results are excluded or given lower weight, creating a quality-differentiated evaluation dataset.
Data Source
AI summary
A model supervision processing method and apparatus, a network-side device, and a readable storage medium. The model supervision processing method of embodiments of this application includes: receiving, by a first network device, first information from a second network device, where the first information includes a data analysis result; and sending, by the first network device, second information to the second network device or a third network device, where the second information is used to indicate a usage status of the data analysis result by the first network device.