Validity Notification for ML Models in Network Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In mobile communication networks, network data analytics functions (NWDAF) face challenges as the data used to train machine learning models may become invalid, leading to incorrect analytics being provided to consumer NFs, potentially causing wrong actions based on outdated information.
Innovation Solution
The NWDAF determines when a trained ML model is no longer valid and notifies relevant NFs, allowing for the re-training of models and providing updated analytics, ensuring that only valid data is used for decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the NWDAF continuously uses the same trained ML model for analytics generation, then the productivity of analytics provision is improved, but the reliability of analytics output deteriorates as the training data becomes invalid
Solution Approach 1:
The system performs preliminary validation of training data before model training to ensure data quality. The NWDAF validates data from producer NFs before using it to train ML models, preventing invalid data from compromising model reliability and eliminating the need for continuous model retraining.
Solution Approach 2:
The system implements a feedback mechanism where consumer NFs provide feedback to the NWDAF about the validity and accuracy of analytics outputs. This feedback loop enables the NWDAF to identify when training data becomes invalid and trigger appropriate actions such as data revalidation or model retraining.
2Reliability
If the NWDAF validates training data from multiple producer NFs before model training, then the reliability of ML model output is improved, but the complexity of data management increases
Solution Approach 1:
The NWDAF performs multiple functions including data collection, validation, model training, and analytics generation within a single network function. This multi-functionality reduces the need for separate validation systems and simplifies the overall architecture while maintaining reliable ML model outputs.
Solution Approach 2:
The NWDAF acts as an intermediary between producer NFs and consumer NFs, managing the validation and processing of training data. By centralizing data validation and model training in the NWDAF, the system reduces the complexity that would otherwise be distributed across multiple NFs.
3Reliability
If the NWDAF frequently re-trains ML models with updated data, then the reliability of analytics is improved, but the energy consumption and processing time increase
Solution Approach 1:
The system performs partial retraining only when necessary, based on validation results and feedback from consumer NFs. Instead of continuous full model retraining, the NWDAF selectively updates models only when training data validity changes, reducing energy consumption while maintaining analytics reliability.
Solution Approach 2:
The system monitors changes in data parameters and model performance metrics to determine when retraining is necessary. By tracking parameter changes in training data from producer NFs, the NWDAF can intelligently decide when to retrain models, avoiding unnecessary processing and energy waste.
Data Source
AI summary
Apparatuses, methods, and systems are disclosed for providing analytics feedback for a network function. One apparatus includes a network interface that receives a first request for a trained machine learning (“ML”) model from a network function (“NF”) that supports analytics generation. Here, the first request contains first information including an Analytic Identifier (“ID”). The apparatus includes processor that determines a first trained ML model based on the first information and provides the first trained ML model to the NF. The processor determines that the first trained ML model is invalid and notifies the NF that the validity of the first trained ML model has changed.