Neural Network Learning Device Dynamic Label Update
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning classifiers for image data often misclassify elements that are originally normal (OK) as abnormal (NG), leading to deteriorated classification accuracy due to the rarity of all NG elements in learning data and reliance on unspecified elements for NG determination.
Innovation Solution
A learning device and method that updates neural network parameters by dynamically adjusting learning reference labels based on statistical tests of NG index values, ensuring accurate classification by distinguishing at least one piece of data belonging to the NG group.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learning data includes multiple elements with OK or NG determination, then machine learning can be performed, but classification accuracy deteriorates because OK elements are misclassified as NG
Solution Approach 1:
The patent segments the learning data into multiple elements (first element, second element, etc.) and processes each element independently through the neural network. By dividing the image data into discrete elements and evaluating them separately, the system can identify which specific elements contain NG characteristics rather than treating the entire image as a single unit, thereby reducing misclassification of OK elements.
Solution Approach 2:
The patent applies local quality by assigning different learning reference labels to different elements based on their individual characteristics. Each element is evaluated with its own reference label (OK or NG), allowing the neural network to learn local abnormalities rather than applying a uniform classification to the entire image. This enables accurate identification of NG elements while preserving OK elements.
2Measurement precision
If NG determination is made based on unspecified elements, then learning can proceed with available data, but classification accuracy deteriorates
Solution Approach 1:
The patent implements feedback by using the neural network's output results to dynamically update learning reference labels. The system evaluates elements using current reference labels, receives output results, and then updates the reference labels based on whether the predictions were correct. This iterative feedback process allows the system to learn from its mistakes and improve classification accuracy even when initial information is incomplete or uncertain.
Solution Approach 2:
The patent performs preliminary action by pre-defining learning reference labels for each element before the learning process begins. These reference labels serve as ground truth that guide the neural network's learning. By establishing these reference labels in advance based on available information, the system can proceed with learning even when complete information about all elements is not initially available.
Data Source
Figure 1
Figure 2
Figure 3~5
AI summary
A learning unit (10) learns a neural network by inputting elements of learning data, on the basis of a learning reference label obtained by assigning an OK label or an NG label to each of the elements of the learning data, and outputs a classifier for determining at least one piece of data belonging to an NG determination group to be NG. A dynamic update unit (11) dynamically updates the learning reference label during learning of the neural network by the learning unit (10).