Base Station AI/ML Model Update Signaling Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In AI/ML global model training with federated learning, the inefficiency of gNB-UE signaling for model update information exchange arises due to the lack of consideration for global model convergence status, leading to increased signaling overhead and model performance degradation from data diversity among user equipment.
Innovation Solution
Enabling or disabling local model update reports from user equipment based on the global model performance status, specifically by a base station considering convergence criteria to manage signaling overhead during the model training phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all user equipments continuously send local model updates to base station, then model training completeness is improved, but signaling overhead increases significantly
Solution Approach 1:
The patent implements dynamic adjustment of model update reporting based on convergence status. The base station monitors whether the global model has converged and dynamically configures UEs to either continuously report updates or stop reporting, transitioning the system from a static reporting mode to a dynamic adaptive mode that responds to training progress
Solution Approach 2:
The patent establishes a feedback mechanism where the base station evaluates convergence status of the global model and sends configuration messages back to UEs. This feedback loop allows the system to adjust reporting behavior based on actual training outcomes, preventing unnecessary signaling when convergence is achieved while ensuring complete data collection during active training phases
2Reliability
If user equipments with diverse local datasets send model updates, then model training coverage is improved, but global model performance degrades due to data diversity
Solution Approach 1:
The patent applies preliminary filtering of UE participation based on convergence assessment. Before allowing UEs with diverse datasets to contribute updates, the system first evaluates whether the global model has converged. This preliminary action prevents harmful diverse updates from being accepted when convergence is near, while still allowing comprehensive data collection during early training stages when diversity beneficially expands training coverage
3Reliability
If model training continues without considering convergence status, then training thoroughness is improved, but communication efficiency deteriorates
Solution Approach 1:
The patent implements convergence-based feedback control where the base station continuously monitors training progress and uses this information to regulate communication between UEs and the network. When convergence is detected, the feedback mechanism switches the communication mode from active update transmission to passive or stopped transmission, thereby maintaining training thoroughness during critical phases while dramatically improving communication efficiency during convergence phases
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a base station, a terminal and a method, a network and a network device, especially cellular mobile telecommunication network element (gNB) and/or user equipment (UE1, UE2, UE3, UE4, ...), - configured to enable or disable by a network device (gNB) considering a global model performance status of machine learning (AI/ML) local model update reports from user equipments (UE1, UE2, UE3, UE4, ...).