Information Processing Apparatus for Label Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning methods face challenges in achieving high prediction accuracy due to errors in labels used for training, leading to inconsistent user determinations and reduced reliability in classification tasks, especially when dealing with data that has continuous variations or is difficult to categorize.
Innovation Solution
An information processing apparatus and method that alternately exchanges a teacher data set and an evaluation data set to improve label consistency, allowing for correction and updating of labels based on prediction results, thereby enhancing the quality of labels used in supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If supervised learning is performed using a data set with labels, then learning can be conducted systematically, but prediction accuracy deteriorates when errors are included in the labels
Solution Approach 1:
The patent implements feedback by using prediction results from the classifier to identify and correct erroneous labels in the data set. The system continuously refines label accuracy by comparing predictions with actual labels and correcting discrepancies, thereby improving prediction accuracy while maintaining learning efficiency
Solution Approach 2:
The patent performs preliminary label correction by using the classifier to predict labels before finalizing the training data. This preliminary action identifies and corrects erroneous labels in advance, ensuring high-quality training data is used for supervised learning
2Reliability
If labels are corrected manually to improve accuracy, then prediction accuracy improves, but processing time and complexity increase
Solution Approach 1:
The patent enables self-service by allowing the system to automatically correct its own labels using the classifier's predictions. The system identifies erroneous labels and corrects them autonomously without requiring manual intervention, thereby improving label accuracy while avoiding time loss
Solution Approach 2:
The system uses feedback from prediction results to automatically identify and correct label errors. This closed-loop process eliminates the need for time-consuming manual correction while maintaining high label accuracy
3Reliability
If manual determination is used to create labels, then label quality can be maintained, but inconsistency arises from different users' determinations
Solution Approach 1:
The patent replaces the mechanical system of manual human determination with an automated classifier-based system. This substitution eliminates inconsistencies arising from different users' subjective judgments while maintaining high label quality through systematic, objective classification
Solution Approach 2:
The system changes the parameter of label creation from human subjective judgment to automated classifier prediction. This parameter change ensures consistent application of classification criteria while maintaining label quality through the classifier's systematic analysis
Data Source
AI summary
Provided is an information processing apparatus including a sorting unit configured to sort a second data set as evaluation data with a classifier generated by learning through supervised learning that uses a first data set as a supervision signal, an input unit configured to receive label correction for the second data set in accordance with a sorting result from the sorting unit, and an update unit configured to update the second data set to reflect the correction received by the input unit.