AI Model Retraining in 6G Wireless Systems

Solution Overview

Problem

In the context of 6G communication systems, there is a need to efficiently retrain artificial intelligence (AI) models used in wireless communication to maintain performance and adapt to changing data distributions, especially when data drift occurs.

Innovation Solution

A method where user equipment (UE) and base stations (BS) collaborate to determine whether to retrain an AI model based on inference information and learning model information, with the UE transmitting a request for retraining to the BS when necessary.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If AI model retraining is performed frequently to maintain performance and adapt to data drift, then model accuracy and adaptability are improved, but system overhead and resource consumption increase

Engineering Contradiction:

Improvemodel performanceVSAvoidsystem overhead

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary monitoring of data distribution characteristics and model performance metrics to detect drift conditions before they significantly degrade model accuracy. By proactively identifying drift through continuous monitoring of statistical properties (e.g., mean, variance) and performance metrics, the system can initiate retraining at optimal moments, preventing performance degradation while avoiding unnecessary retraining operations that would increase system overhead.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where model performance metrics and data distribution characteristics are continuously monitored and fed back to the drift detection module. This feedback loop enables the system to dynamically adjust retraining timing based on actual performance degradation patterns, ensuring retraining is performed only when necessary to maintain accuracy while minimizing unnecessary retraining operations that would consume resources.

Inventive Principle:

Principle #23Feedback

2Device complexity

If AI model retraining is delayed to reduce system overhead, then resource consumption is reduced, but model accuracy and adaptability deteriorate

Engineering Contradiction:

Improvesystem overheadVSAvoidmodel performance

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The system performs preliminary monitoring of data distribution characteristics and model performance metrics to detect drift conditions before they significantly degrade model accuracy. By proactively identifying drift through continuous monitoring of statistical properties (e.g., mean, variance) and performance metrics, the system can initiate retraining at optimal moments, preventing performance degradation while avoiding unnecessary retraining operations that would increase system overhead.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where model performance metrics and data distribution characteristics are continuously monitored and fed back to the drift detection module. This feedback loop enables the system to dynamically adjust retraining timing based on actual performance degradation patterns, ensuring retraining is performed only when necessary to maintain accuracy while minimizing unnecessary retraining operations that would consume resources.

Inventive Principle:

Principle #23Feedback

3Loss of time

If comprehensive monitoring of data drift is implemented to detect drift early, then retraining timing is optimized, but computational overhead increases

Engineering Contradiction:

Improveretraining timingVSAvoidcomputational overhead

Core Design Contradiction:

Loss of timeVSUse of energy by moving object

Solution Approach 1:

The system applies partial monitoring by focusing on critical subsets of data characteristics and performance metrics rather than comprehensively analyzing all data. By monitoring key statistical properties (mean, variance) of data distributions and selected performance metrics, the system detects drift with sufficient accuracy to trigger retraining while significantly reducing computational overhead compared to exhaustive monitoring approaches.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system performs preliminary monitoring of data distribution characteristics and model performance metrics to detect drift conditions before they significantly degrade model accuracy. By proactively identifying drift through continuous monitoring of statistical properties (e.g., mean, variance) and performance metrics, the system can initiate retraining at optimal moments, preventing performance degradation while avoiding unnecessary retraining operations that would increase system overhead.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250200385A1Method and apparatus for performing retraining of artificial intelligence model in wireless communication system

Publication Date: 2025.06.19 SAMSUNG ELECTRONICS CO LTD

AI summary

Provided is a 5th-generation (5G) or 6th-generation (6G) communication system for supporting higher data rates after the 4th-generation (4G) communication system such as long term evolution (LTE). A method by which a user equipment (UE) performs communication includes receiving, from a base station (BS), learning model information for an artificial intelligence (AI) model. The method includes determining whether to retrain the AI model, based on inference information obtained by using the AI model and the learning model information. The method includes transmitting, to the BS, a request message for retraining the AI model, in case that the retraining of the AI model is determined.