Inference-Aware ML Model Provisioning in 5G Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ML model provisioning in 5G/NR systems often results in excessive computational and networking overhead without significant accuracy improvements, as the decision-making process lacks sufficient information about inference-related factors, leading to inefficient resource utilization and suboptimal model provisioning.
Innovation Solution
The method involves obtaining and utilizing machine-learning model request information that includes both model-related and inference-related data, such as accuracy requirements, data sources, and execution environment details, to make informed decisions on selecting, modifying, or generating ML models, thereby optimizing the provisioning process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML model re-training or new model generation is performed to improve accuracy, then model accuracy is improved, but computational load and networking overhead increase significantly
Solution Approach 1:
The patent changes the parameter set used for decision-making by the MTLF. Instead of relying only on local unspecified logic, the system now considers inference-related information parameters (inference accuracy requirements, data sources, execution environment) provided by the service consumer. This expanded parameter set enables the MTLF to select operations that balance accuracy improvements with computational efficiency, avoiding unnecessary re-training when existing models suffice.
Solution Approach 2:
The patent implements a feedback mechanism where the service consumer provides inference-related information back to the MTLF. This feedback loop allows the MTLF to understand the actual inference requirements and adjust its model provisioning decisions accordingly. The MTLF can now receive feedback about accuracy requirements and execution environments, enabling it to make informed decisions about whether re-training is truly necessary or if existing models can meet the requirements.
2Measurement precision
If ML model re-training or new model generation is performed to improve accuracy, then model accuracy is improved, but networking overhead increases due to additional data collection
Solution Approach 1:
The patent introduces inference-related information parameters that change the decision-making parameters of the MTLF. By considering data source requirements and execution environment information provided by the service consumer, the MTLF can determine whether existing models already meet the needs or if new data collection and re-training are truly necessary. This parameter expansion prevents unnecessary networking overhead for data collection when existing models suffice.
3Ease of operation
If the MTLF uses local unspecified logic for model provisioning decisions, then decision-making is simple, but resource utilization becomes inefficient
Solution Approach 1:
The patent makes the MTLF serve multiple functions: it not only selects models based on accuracy requirements but also optimizes resource utilization by considering inference-related information. The MTLF becomes a multi-functional entity that balances accuracy improvements with computational and networking efficiency, adapting its operations based on the specific inference requirements provided by service consumers.
4Use of energy by moving object
If existing trained ML models are selected without further processing, then computational load is minimized, but model accuracy may be insufficient
Solution Approach 1:
The patent implements feedback from the service consumer to the MTLF regarding inference accuracy requirements and execution environment. This feedback enables the MTLF to determine when existing models are sufficient and when re-training or new model generation is truly necessary. The MTLF can now make informed decisions by comparing existing model capabilities against the actual inference requirements provided through the feedback mechanism.
Data Source
AI summary
There are provided measures for enabling/realizing inference-aware ML (machine learning) model provisioning, e.g. to support network data analytics, in a mobile/wireless communication system. Such measures exemplarily comprise that ML model request information, including model-related information indicating one or more properties of a requested ML model and inference-related information indicating one or more properties of execution of inference based on the requested ML model, is provided from a first network entity (representing a service consumer of a network data analytics service) to a second network entity (representing a service provider of the network data analytics service), the second network entity specifies an ML model to be provisioned based on the ML model request information, and ML model information about the specified ML model is provided from the second network entity to the first network entity.