Adaptive AI/ML Model Updating for Multi-Scenario Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML models used for terminal device location estimation suffer significant accuracy drops when applied to scenarios different from those in which they were trained, necessitating an effective method for model updating.
Innovation Solution
A method for determining the target updating mode of a model based on performance parameters and indication information, including updating model parameters, transitioning to a new model, or training a new model in multiple scenarios to maintain accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an AI/ML model is trained in a specific scenario and applied to other scenarios, then the model can be reused without retraining, but positioning accuracy drops significantly
Solution Approach 1:
The patent implements dynamic model updating mechanisms that allow the model to adapt to different scenarios. The system determines whether to update model parameters, update the entire model, or retrain based on scenario changes, making the model flexible and adaptable while maintaining positioning accuracy across diverse environments
Solution Approach 2:
The patent changes model parameters adaptively based on scenario requirements. By adjusting model parameters according to the specific scenario being applied to, the system maintains positioning accuracy while enabling model reuse across different contexts without complete retraining
2Measurement precision
If model parameters and structure are updated to maintain accuracy in new scenarios, then positioning accuracy is preserved, but computational complexity and training requirements increase
Solution Approach 1:
The patent applies partial updating by selectively updating only certain model parameters or components rather than the entire model. This partial action approach maintains positioning accuracy while reducing computational complexity and training requirements compared to complete model updates
Solution Approach 2:
The patent segments the model updating process into different levels: parameter updates, model structure updates, and complete retraining. This segmentation allows the system to choose the appropriate level of updating based on scenario requirements, reducing unnecessary computational complexity while maintaining accuracy
3Measurement precision
If a new model is trained based on sample sets in multiple scenarios, then positioning accuracy across different scenarios improves, but training time and data requirements increase
Solution Approach 1:
The patent creates a universal model that can function across multiple scenarios by training on diverse sample sets. This multi-functional model reduces the need for scenario-specific models and their associated training times, as one model serves multiple purposes across different environments
Solution Approach 2:
The patent performs preliminary training on diverse sample sets covering multiple scenarios to create a robust base model. This preliminary action prepares the model to handle various scenarios without requiring extensive retraining later, reducing overall training time while maintaining multi-scenario positioning accuracy
Data Source
AI summary
A model updating method includes: determining, by a first device, a target updating mode of a first model in a plurality of updating modes based on first information and/or first indication information of a second device; where the first model is used to determine location related information of a terminal device, the first information comprises performance parameter(s) of a model, the first indication information is used to indicate the target updating mode, and the plurality of updating modes include at least two of: a first updating mode, indicating that a part of model parameters of the first model and/or a model structure of the first model is updated; a second updating mode, indicating that the first model is updated to a second model; and a third updating mode, indicating that the first model is updated to a third model.