Model Training Logical Function Accuracy Monitoring in 5G Core

The patent segments the NWDAF into two separate logical functions: AnLF for analytics inference and MTLF for model training and accuracy monitoring. This segmentation allows MTLF, which has greater computational and storage capabilities, to handle model retraining and accuracy monitoring tasks that would be too complex for the lightweight AnLF.

The patent introduces an intermediary mechanism where MTLF acts as a supporting function for AnLF. MTLF receives requests from AnLF for model retraining and accuracy monitoring, processes these requests using its greater computational resources, and returns results to AnLF. This intermediary relationship allows AnLF to benefit from enhanced accuracy capabilities without bearing the computational burden.

The patent extracts the accuracy monitoring and comparison functionality from the core AnLF and places it within MTLF. MTLF is responsible for retrieving actual outcomes, comparing them with predictions, calculating accuracy metrics, and determining when retraining is needed. This extraction reduces the complexity burden on AnLF while maintaining comprehensive accuracy monitoring capabilities.

The patent implements a feedback mechanism where MTLF continuously monitors analytics accuracy by comparing predictions with actual outcomes. When accuracy falls below thresholds or retraining intervals are reached, MTLF triggers model retraining and provides updated models to AnLF. This closed-loop feedback system enables continuous accuracy improvement without requiring AnLF to directly implement complex monitoring logic.