Network ML Model Training via Capability-Aware Control Signaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of operator networks due to increased terminal devices and diversified services complicates network operation and maintenance, and real-time changes in network or service performance requirements make it challenging to ensure effective network or service performance using existing automation or intelligence technologies.
Innovation Solution
A communication method and apparatus that enables autonomous management of machine learning (ML) models by exchanging capability and control information between devices, allowing devices to train ML models autonomously and ensure network or service performance meets current requirements, avoiding failures due to capability mismatches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the operator network management device delivers the MnF model to the network element device through multiple interactions, then the network management process becomes complex and time-consuming, but the network or service performance requirements change in real time making the delivered model ineffective
Solution Approach 1:
The network element device is equipped with autonomous model training capabilities, allowing it to independently train and update ML models based on real-time network data without requiring continuous operator intervention. The device can autonomously determine when and how to train models, reducing management complexity while ensuring performance requirements are met
Solution Approach 2:
The system enables dynamic model training where the network element device can adaptively update ML models in real-time based on changing network conditions and performance requirements. This dynamic capability allows the model to evolve with network demands rather than remaining static after initial delivery
2Ease of operation
If the operator network management device delivers a fixed MnF model to the network element device, then the management process is simplified, but the model cannot meet real-time changing network or service performance requirements
Solution Approach 1:
The network element device is pre-equipped with model training capabilities and necessary algorithms before deployment. This preliminary preparation enables the device to autonomously adapt to changing requirements without requiring complex real-time reconfiguration or redelivery of models by the operator
Solution Approach 2:
The system enables self-updating of ML models where the network element device automatically retrains models based on new data and changing performance requirements, eliminating the need for manual model redelivery while maintaining adaptability
3Reliability
If the control information does not match the capability information of the first device, then the model training may fail, but ensuring compatibility reduces the efficiency of deploying ML models
Solution Approach 1:
The system implements a feedback mechanism where the network element device reports its capability information to the operator network management device, which then generates control information that is validated against these capabilities before deployment. This feedback loop ensures compatibility while maintaining deployment efficiency through automated validation
Solution Approach 2:
The capability information exchange and control information generation are performed in advance before actual model training begins. This preliminary matching prevents training failures by ensuring compatibility is established beforehand, avoiding the need for retraining or reconfiguration
Data Source
AI summary
A communication method and apparatus. A first device sends capability information to a second device, so that the second device can send control information to the first device based on the capability information. The control information is usable to indicate that a first machine learning (ML) model corresponding to a first management function (MnF) on the first device is allowed to be trained. The first device trains the first ML model based on the control information.