Feature-Target Data Display for Efficient ML Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of automating labeling in machine learning is significant, as users struggle to recognize the correspondence between target data and features, leading to inefficiencies in the labeling process.
Innovation Solution
An information processing device and method that facilitate the display of target data and features in a manner identifiable to users, allowing for intuitive selection and labeling of features through a user-friendly interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If features are extracted from target data for machine learning, then the learning model accuracy is improved, but the user's ability to recognize the correspondence between target data and features deteriorates
Solution Approach 1:
The patent introduces an intermediary display system that shows both the extracted features and the original target data simultaneously. This intermediary interface allows users to verify the correspondence between features and target data without affecting the feature extraction process itself, thus maintaining learning model accuracy while improving user recognition capability.
Solution Approach 2:
The patent creates visual copies or representations of both the target data and extracted features in the display interface. By presenting replicated views of the original data alongside the processed features, users can directly compare and verify the correspondence relationship without altering the actual feature extraction or model training process.
2Measurement precision
If manual labeling is performed to ensure data quality, then the labeling accuracy is improved, but the labeling efficiency deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the display interface shows users the correspondence between target data and extracted features, enabling them to verify and correct labeling decisions. This feedback loop allows users to maintain high labeling accuracy while working more efficiently by providing immediate visual confirmation of feature-target relationships.
Solution Approach 2:
The patent performs preliminary feature extraction and display preparation before the actual labeling task. By pre-processing and presenting the correspondence information in an easily visible format, users can perform labeling more efficiently without compromising accuracy, as the difficult analytical work of feature extraction has already been completed automatically.
3Productivity
If features are used instead of displaying target data, then data processing efficiency is improved, but the user's understanding of actual data deteriorates
Solution Approach 1:
The patent merges the display of extracted features with the original target data in a unified interface. This combination allows the system to maintain the processing efficiency benefits of feature extraction while simultaneously preserving the user's understanding of the actual data by showing both representations together in context.
Solution Approach 2:
The patent adds a new display dimension that simultaneously presents both the processed features and the original target data. By utilizing this additional dimensional space in the user interface, the system avoids the trade-off between processing efficiency and information preservation, allowing users to access both representations without increasing processing complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided are an information processing device, a program and a method that enable a user to conveniently check the correspondence between target data and a feature based on the target data. A program causes a computer to execute acquiring a plurality of pieces of target data, acquiring a plurality of features based on the plurality of pieces of target data, receiving selection of at least one feature among the plurality of features, and generating a display screen including a target data display area in which the plurality of pieces of target data are displayed and a feature display area in which the plurality of features are displayed, the display screen displaying a set of the at least one selected feature and at least associated one of the plurality of pieces of target data in a manner identifiable to a user.