5G Core ML Model Management for Multi-Model Analytics Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 5G networks face challenges in managing multiple machine learning (ML) models used for generating analytic reports, leading to ambiguities and difficulties in managing ML models and the analytics they produce, due to the implicit assumption of a one-to-one relationship between ML models and Analytics IDs.
Innovation Solution
Introduce explicit references to ML models and conditions for training and swapping ML models in the Network Data Analytics Function (NWDAF), using improved techniques for ML model management, including enhanced Nnwdaf_MLModelProvision and Nnwdaf_MLModelInfo services to facilitate better management of ML models for analytics in 5G networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple ML models are used to generate analytic reports for a particular Analytics ID, then the versatility and accuracy of analytics are improved, but the complexity of managing ML models and ambiguities in model-analytics relationships increase
Solution Approach 1:
The patent segments the management of ML models by introducing distinct logical functions (MTLF for model training and AnLF for analytics) within NWDAF. This segmentation allows independent management of model lifecycle (training, validation, storage) separate from analytics generation, reducing management complexity while supporting multiple models per Analytics ID through structured identification mechanisms.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of enhanced service interfaces (Nnwdaf_MLModelProvision, Nnwdaf_MLModelInfo) that mediate between model training and analytics consumption. These services act as intermediaries that manage the many-to-many relationships between ML models and Analytics IDs, providing structured model identification and version tracking to reduce ambiguity.
2Ease of operation
If explicit references to ML models and conditions for training and swapping models are introduced in NWDAF, then the ease of operation and model management are improved, but the device complexity and service structure complexity increase
Solution Approach 1:
The patent applies preliminary action by establishing model training and validation processes (MTLF) before models are deployed for analytics (AnLF). The Nnwdaf_MLModelProvision service performs preliminary model provisioning, training, and validation, storing model artifacts and metadata in advance. This preliminary structuring simplifies subsequent model swapping and analytics generation operations.
Solution Approach 2:
The patent introduces dynamic model swapping capabilities where AnLF can switch between different ML models based on performance conditions, availability, or analytics requirements. The system dynamically manages model versions and can retrieve different model artifacts as needed, providing flexibility without requiring complete service restructuring.
3Productivity
If enhanced Nnwdaf_MLModelProvision and Nnwdaf_MLModelInfo services are implemented, then the productivity and efficiency of ML model management are improved, but the difficulty of detecting and measuring model states and relationships increases
Solution Approach 1:
The patent implements feedback mechanisms where Nnwdaf_MLModelInfo service provides detailed model information including version, status, and performance metrics back to the system. The Nnwdaf_MLModelProvision service receives feedback on model training completion and validation results. This structured feedback enables efficient model state tracking and management despite multiple models and versions.
Solution Approach 2:
The patent uses copying by creating and storing metadata copies of ML models including model identifiers, version information, and artifact locations. The Nnwdaf_MLModelProvision service stores model metadata and artifacts separately from the analytics processing, enabling efficient reference and tracking without duplicating the entire model management burden in the analytics path.
Data Source
AI summary
Embodiments include methods for a first network node or function (NNF) configured for machine learning (ML) model management in a communication network. Such methods include receiving, from a second NNF of the communication network, a first message including: one or more ML model identifiers corresponding to one or more ML models maintained by the first node, or an identifier of an analytic based on the ML model(s). Such methods include sending, to the second NNF, a second message including one of the following: a plurality of tuples corresponding to a plurality of ML models on which the analytic is based, each tuple including a different ML model identifier and information element(s) associated with the corresponding ML model; or a single tuple including the analytic identifier and information element(s) associated with a single ML model on which the analytic is based. Other embodiments include complementary methods for the second NNF.