AI/ML Model Renewal for Network Context Adaptation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Machine learning models used in user equipment (UE) and network functions (NF) become unsuitable when the UE moves away from the network where the models were trained, leading to performance degradation due to changes in network context, with no existing mechanisms to detect or address this mismatch.

Innovation Solution

Implement mechanisms to detect mismatches between pre-trained AI/ML models and current network contexts, allowing for model renewal by retraining the models with new data from the current context, using AI/ML monitoring models to profile datasets and evaluate performance, and updating models when performance thresholds are exceeded.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a pre-trained AI/ML model is deployed in a UE, then the model can provide accurate predictions and actions for the network context where it was trained, but the model performance degrades when the UE moves to a different network with different measurement data characteristics

Engineering Contradiction:
Improvemodel prediction accuracyVSAvoidmodel adaptability to different network contexts
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic model adaptation by enabling UEs to detect when network context changes occur and automatically request updated AI/ML models from the network. The system transitions from a static pre-trained model to a dynamic system where models can be renewed based on detected context changes, ensuring continued accuracy across different network environments

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent establishes a feedback mechanism where the UE monitors network measurement data characteristics, compares them against the training data profile, and triggers model renewal requests when mismatches are detected. This closed-loop feedback ensures the model remains adapted to current network conditions

Inventive Principle:
Principle #23Feedback

2Reliability

If the UE continuously monitors network context changes to detect model mismatch, then the model can be renewed timely to maintain performance, but the complexity of the system increases due to additional monitoring and detection mechanisms

Engineering Contradiction:
Improvemodel performance maintenanceVSAvoidsystem complexity for model monitoring
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements self-service by enabling the UE to autonomously monitor its own network context, detect mismatches between current and training data characteristics, and automatically request model renewals without requiring complex external monitoring infrastructure. The UE serves itself in detecting when model renewal is needed

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent introduces an intermediary mechanism where the UE acts as a mediator between the network and the AI/ML model lifecycle. The UE monitors context changes and communicates renewal needs to the network, simplifying the overall system architecture by distributing the monitoring function to the edge device

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4285519B1Machine learning model renewal
Publication Date: 2024.11.27 NOKIA TECHNOLOGIES OY
  • EP4285519B1 patent drawingFigure 1
  • EP4285519B1 patent drawingFigure 2
  • EP4285519B1 patent drawingFigure 3

AI summary

There are provided measures for machine learning model renewal. Such measures exemplarily comprise receiving a first machine learning model message including a first machine learning inference model, obtaining network related input data, feeding said first machine learning inference model with said network related input data, receiving, upon unsuitability of said first machine learning inference model for an experienced network condition, a second machine learning model message including a second machine learning inference model, and replacing said first machine learning inference model with said second machine learning inference model.