Positioning Training Sample Screening for Noisy Label Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML models for positioning in telecommunication systems face challenges in accurately evaluating training samples with noisy labels, which affects the model's training efficiency and positioning accuracy.
Innovation Solution
A device evaluates the quality of training samples by determining a quality parameter based on target accuracy and label information, transmitting a report to another device to improve the quality of training samples before they are used for model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training samples with noisy labels are used for AI/ML model training, then the quantity of training data increases, but the positioning accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by evaluating the quality of training samples before they are used for model training. The quality evaluation process assesses label quality metrics and determines whether samples meet predefined quality thresholds, filtering out noisy samples in advance to prevent them from degrading model training performance
Solution Approach 2:
The patent extracts and removes low-quality training samples from the dataset through quality evaluation. By identifying samples with noisy labels using quality metrics and thresholds, the system separates useful training data from harmful noisy data, ensuring only high-quality samples are used for training
2Reliability
If all training samples are transmitted for evaluation, then the evaluation completeness improves, but the transmission overhead increases
Solution Approach 1:
The patent performs preliminary quality evaluation on training samples before transmission to the model training system. By assessing label quality and determining sample suitability in advance, the system avoids transmitting unnecessary low-quality samples, reducing transmission overhead while maintaining evaluation completeness for samples that meet quality criteria
Data Source
AI summary
Example embodiments of the present disclosure relate to positioning enhancements. A first device receives, from a second device, first information indicating a target accuracy for positioning and a set of parameters for a label quality evaluation of a training sample. The training sample comprises a radio measurement and label information associated with the radio measurement. The first device determines a quality parameter of the training sample based on the set of parameters, the label information. The first device then transmits a report at least comprising the quality parameter to the second device. In this way, a model for positioning can be well trained with the evaluated training sample, and thus the positioning accuracy is improved.