User Interface for Unbiased Data Division in Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing sites, the challenge is to divide data for machine learning into learning and evaluation sets without bias, especially in environments where data acquisition is limited, such as factories, where manual operations require significant man-hours and multidimensional data handling is complex, leading to potential data bias.
Innovation Solution
A method for displaying data groups statistically and visually to facilitate unbiased division, using a user interface unit and display device that sorts, filters, and presents data in a format allowing easy recognition of biases, enabling users to adjust and divide data without bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual division of data into learning and evaluation sets is performed, then data can be divided without bias, but a large number of man-hours are required
Solution Approach 1:
The patent introduces a user interface unit as an intermediary tool between the raw data and the final divided datasets. This interface automatically performs statistical analysis and visualizes data distribution, allowing operators to review and adjust divisions with minimal manual effort while maintaining high data quality
Solution Approach 2:
The patent replaces manual mechanical data division operations with automated computational systems that perform statistical analysis and visualization, substituting human labor with algorithmic processing while preserving operator oversight capability
2Productivity
If simple random division method is used, then data division is fast and easy, but bias occurs between learning and evaluation data in limited factory environments
Solution Approach 1:
The patent implements feedback mechanisms where the system automatically analyzes data distribution statistics and visualizes them for operators. This feedback loop allows operators to identify potential biases and adjust the division accordingly, ensuring both efficiency and quality
Solution Approach 2:
The patent performs preliminary statistical analysis and visualization before final data division is committed. This preliminary action allows operators to detect and correct potential biases early in the process, preventing biased divisions from being finalized
Data Source
AI summary
An inspection apparatus includes a group data creation unit configured to create group data obtained by dividing a plurality of samples into a plurality of groups, a statistical state calculation unit configured to perform a statistical process for the samples divided into each of the groups to calculate data indicating a statistical state of a predetermined data item between the groups, and a user interface unit configured to display the statistical state in a display format enabling recognition of the statistical state between the groups on the basis of the data indicating the statistical state of the predetermined data item between the groups calculated by the statistical state calculation unit.