Semi-Supervised Training Dataset Generation for Wall Diagnostics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating training datasets for wall diagnostic devices lack precision and reliability in classifying sensor data, particularly for objects within walls, leading to inadequate performance of artificial intelligence in wall diagnostics.
Innovation Solution
A method involving a classifier module trained via semi-supervised learning to classify unclassified sensor data using classified sensor data and pseudo-classified data, and employing latent space representations and distance determination to enhance classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional supervised learning is used to train the classifier module, then the training dataset requires extensive manual classification effort and time, but the classification precision and reliability remain insufficient
Solution Approach 1:
The system employs semi-supervised learning where the classifier module progressively improves its own classification capabilities by learning from initially classified data and progressively incorporating pseudo-classified data generated by itself, reducing dependency on continuous manual annotation while improving classification precision over time
Solution Approach 2:
A small initial set of manually classified sensor data is prepared in advance to initialize the classifier module, enabling it to start generating pseudo-classified data before full-scale operation, thus reducing the overall time required for training dataset generation
2Reliability
If more manually classified sensor data is collected to improve AI performance, then the classification reliability improves, but the data collection and labeling effort increases significantly
Solution Approach 1:
The classifier module serves itself by generating pseudo-classified data from unclassified sensor data, creating its own training material without requiring external manual annotation efforts, thus improving reliability without increasing data collection complexity
Solution Approach 2:
Pseudo-classified data acts as an intermediary between manually classified data and fully automated classification, bridging the gap by providing intermediate training samples that help the classifier progressively improve its reliability without direct human involvement
Data Source
AI summary
A computer-implemented method for generating a training dataset for training an artificial intelligence for operating a measuring device, in particular a wall diagnostic device, includes (i) receiving unclassified sensor data of at least one sensor unit of a measuring device by a classifier module, (ii) classifying the unclassified sensor data and providing classified sensor data using a classifier module, and (iii) adding the classified sensor data to a training dataset.