Machine Learning Model Retraining via Pseudo Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models experience accuracy decreases during system operation due to changes in operation data features, such as camera contamination or environmental changes, making it challenging to maintain classification accuracy in tasks like semantic segmentation without manual intervention or pre-prepared training data.
Innovation Solution
A machine learning apparatus generates synthetic pseudo labels for images with good classification results and uses these to create training data for retraining the model, labeling images with poor classification results, thereby maintaining model accuracy without manual labeling or pre-prepared data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the machine learning model is trained using manually labeled training data to maintain accuracy, then classification accuracy is improved, but operational cost and time consumption increase
Solution Approach 1:
The system performs self-labeling by automatically generating labels for operation data using the machine learning model itself. The model classifies images during operation, and these classification results are used as labels for retraining, eliminating the need for manual human labeling while maintaining accuracy
Solution Approach 2:
The system accumulates and stores classified images with their classification results during operation before retraining. This preliminary accumulation of labeled data during the operation period prepares the training dataset in advance, reducing the need for time-consuming manual labeling when retraining is needed
2Measurement precision
If the machine learning model is retrained frequently to adapt to changing operation data, then classification accuracy is improved, but computational cost and processing time increase
Solution Approach 1:
The system performs retraining at predetermined intervals rather than continuously or frequently. The classification accuracy is evaluated periodically, and retraining is triggered only when necessary or at scheduled times, reducing unnecessary computational overhead while maintaining accuracy
Solution Approach 2:
The system uses a partial retraining approach by selecting only certain images (e.g., a predetermined number of images or images meeting specific criteria) for retraining instead of using all accumulated data, reducing the computational burden while still improving accuracy
3Measurement precision
If manual confirmation of classification results is performed to ensure accuracy, then determination accuracy is improved, but operational complexity and cost increase
Solution Approach 1:
The system implements an automated feedback mechanism where classification results are automatically evaluated against accuracy thresholds. When accuracy drops below the threshold, the system automatically triggers retraining without requiring manual confirmation, reducing operational complexity while maintaining accuracy
Solution Approach 2:
The machine learning model performs self-evaluation of its own classification accuracy and self-triggered retraining when needed. This automated self-monitoring and self-correction eliminates the need for manual confirmation operations, reducing both complexity and cost
Data Source
AI summary
A non-transitory computer-readable recording medium stores a machine learning program for causing a computer to execute processing including: in a case where a machine learning model classifies a first image based on a value less than a threshold, generating training data in which a classification result of a second region of a second image that corresponds to a position of a first region of the first image is labeled to the first region, based on a classification result obtained by classifying the second image based on a value equal to or more than the threshold by the machine learning model; and training the machine learning model based on the training data.