Information Processing Device for Machine Learning Data Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for improving machine learning efficiency by calculating similarity degrees between images and extracting learning data based on these similarities fail to provide comprehensive information, particularly in identifying the cause of execution results issues with learned models.
Innovation Solution
An information processing device with an acquisition unit to acquire input data, an extraction unit to extract related input data based on similarity degrees with past input history data, and a presentation unit to present this extracted data, enabling the presentation of useful information to users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional techniques calculate similarity degrees between images and extract learning data based on these similarities, then learning efficiency is improved, but the ability to provide comprehensive information about execution results issues is insufficient
Solution Approach 1:
The patent segments the information presentation into multiple dimensions: similarity-based extraction and cause-category-based presentation. By dividing the analysis into similarity degree calculation and cause classification, the system maintains learning efficiency while providing comprehensive information about execution results issues through structured cause categories (data quality, model architecture, hyperparameters, environment).
Solution Approach 2:
The patent introduces cause categories as an intermediary layer between the similarity extraction process and the final information presentation. This intermediary structure organizes the extracted learning data according to predefined cause categories, enabling comprehensive information delivery without compromising the efficiency of the similarity-based extraction mechanism.
2Speed
If conventional techniques present particular images using similarity degrees, then processing speed is maintained, but the ability to identify causes of execution results issues is limited
Solution Approach 1:
The patent segments the information presentation by introducing cause categories that classify potential issues (data quality, model architecture, hyperparameters, environment). This segmentation allows the system to maintain fast similarity-based processing while systematically organizing results to facilitate cause identification, making it easier to detect and measure execution problems.
Solution Approach 2:
The patent performs preliminary action by pre-defining cause categories before the similarity extraction process. These predefined categories serve as a prepared framework that guides the organization and presentation of extracted learning data, enabling rapid cause identification without adding processing overhead during the similarity calculation phase.
3Measurement precision
If the system extracts learning data based on similarity degrees, then relevant information is obtained, but structured presentation for various situations is insufficient
Solution Approach 1:
The patent applies universality by designing a cause category framework that can handle multiple types of execution results issues through a unified structure. The same similarity-based extraction mechanism serves multiple purposes: retrieving relevant learning data and organizing it according to different cause categories (data quality, model architecture, hyperparameters, environment), thereby providing adaptable presentation for various situations without sacrificing information relevance.
Data Source
AI summary
There is provided an information processing device that includes an acquisition unit configured to acquire first input data input when first output data is obtained in predetermined processing of obtaining output data with respect to input data, an extraction unit configured to extract second input data related to the first input data acquired by the acquisition unit based on a similarity degree between the first input data and each input history data, which is a history of input data of a case of past execution of the predetermined processing, from the input history data, and a presentation unit configured to present the second input data extracted by the extraction unit.