ML Model Context Tagging for Network-State Analytics Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to account for the network state during which machine learning models were trained, leading to suboptimal selection and utilization of ML models in network analytics, as the same model trained under different network conditions can produce distinct results.
Innovation Solution
Introduce a network state parameter to describe the training conditions of ML models, enabling efficient storage, search, and retrieval of ML models in the ADRF, allowing consumers to select models based on the current network state.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML models are stored without network state parameters, then storage simplicity is maintained, but model selection accuracy deteriorates
Solution Approach 1:
The network state parameter is nested within the ML model data structure as an associated attribute. When an ML model is stored in the ADRF, the corresponding network state parameter is simultaneously stored and linked to it, creating a nested relationship where the parameter is contained within the model's metadata structure. This allows the system to maintain a unified storage mechanism while capturing additional contextual information.
Solution Approach 2:
The network state parameter acts as an intermediary between the ML model and the selection process. Instead of directly comparing models based on performance metrics alone, the system uses the network state parameter as a mediating factor that bridges the gap between model characteristics and deployment conditions, enabling more accurate model selection by matching both the model capabilities and the current network environment.
2Reliability
If network state parameters are included with ML models, then analytics relevance is improved, but information processing complexity increases
Solution Approach 1:
The network state parameter is determined and recorded at the time of ML model training, before the model is deployed or selected for inference. This preliminary capture of contextual information ensures that when the model is later retrieved from the ADRF, the associated network state is already available, eliminating the need for real-time state assessment during model selection and reducing processing complexity at inference time.
3Reliability
If ML models are selected based on training conditions, then model performance is improved, but selection process complexity increases
Solution Approach 1:
The ML model selection process is enhanced to automatically utilize the network state parameter that is already associated with each stored model. When a model selection request is made, the system self-service by comparing the current network state with the stored parameters of available models, automatically identifying and selecting the most appropriate model without requiring manual intervention or complex external evaluation processes.
Data Source
AI summary
There is provided an apparatus comprising and a processor coupled with a memory. The processor is configured to cause the apparatus to receive a request message for an analytics service that uses a machine learning (ML) model, the request comprising a use case parameter; determine the ML model based on the use case parameter and a current network state, wherein the ML model is for deriving analytics information for a wireless communication network; and transmit a response message comprising analytics service information based on the determined ML model.