Machine Learning Apparatus Identifying Inappropriate Data Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques for supervised learning are inefficient in identifying training data with inappropriate labels due to the reliance on test data that is also used for learning, leading to low accuracy in specifying such data from large datasets.
Innovation Solution
A machine learning apparatus and method that generates multiple learning groups from training data, learns distinct data determination models for each group, and uses external target data to identify inappropriate labels by comparing results across models, allowing for efficient specification of data with potentially incorrect labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If training data is used as test data to identify inappropriate labels, then the process is simple, but the efficiency of specifying inappropriate data is low
Solution Approach 1:
The training data set is divided into multiple learning groups where each group contains at least one piece of data that is different from the others. This segmentation allows the system to train multiple determination models on different data subsets, enabling more effective identification of inappropriate labels by comparing results across models trained on varied data compositions
2Productivity
If multiple learning groups and determination models are used to improve identification accuracy, then the efficiency of specifying inappropriate data improves, but the device complexity increases
Solution Approach 1:
The training data set is divided into multiple learning groups where each group contains at least one piece of data that is different from the others. This segmentation allows the system to train multiple determination models on different data subsets, enabling more effective identification of inappropriate labels by comparing results across models trained on varied data compositions
Solution Approach 2:
Multiple determination models serve the universal function of identifying inappropriate labels in training data. Each model, trained on a different learning group, performs the same classification task but provides diverse perspectives that collectively improve the accuracy of detecting mislabeled data
Data Source
AI summary
A machine learning apparatus (1) includes: a storage unit (11) configured to store a training data set; a group generation unit (12) configured to generate, from the training data set, a plurality of learning groups; a learning unit (13) configured to learn a predetermined data determination model for each of the plurality of learning groups and store a plurality of learned models that correspond to the respective learning groups in the storage unit (11); a determination unit (14) configured to input, to each of the plurality of learned models, determination target data that is not included in the corresponding learning group, and acquire first determination results; and a specifying unit (15) configured to specify a first inappropriate data group in which the label is possibly inappropriate from the training data set based on the first determination results and a correct answer label in the determination target data.