Dynamic Slicing of Machine Learning Programs via Target Data Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Dynamic slicing of machine learning programs is inefficient due to the large amount of data used, leading to prolonged execution times, and reducing training data can hinder execution reliability.

Innovation Solution

A method to reduce the data amount for dynamic slicing by extracting and outputting sampling data that includes only the target data necessary for the machine learning program, determined through operation descriptions and extraction conditions, allowing for efficient execution while maintaining reliability.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the full training data is used for dynamic slicing, then the execution reliability is maintained, but the execution time becomes excessively long

Engineering Contradiction:

Improveexecution reliabilityVSAvoidexecution time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the full training data into multiple subsets (first training data and second training data) based on the divided code snippets. This allows dynamic slicing to be performed on smaller, manageable portions of data rather than the complete dataset, significantly reducing execution time while maintaining the ability to process the full dataset through multiple iterations

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary actions by dividing the machine learning program into multiple code snippets before dynamic slicing. This pre-processing step enables the system to prepare and organize data subsets in advance, allowing the dynamic slicing process to operate efficiently on pre-segmented data rather than processing the entire dataset at once

Inventive Principle:

Principle #10Preliminary action

2Productivity

If the training data amount is reduced for faster execution, then the execution time is shortened, but the execution reliability may be compromised

Engineering Contradiction:

Improveexecution efficiencyVSAvoidexecution reliability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent ensures continuity of useful action by iteratively processing multiple data subsets through the dynamic slicing process. Instead of performing a single pass on reduced data, the system continuously processes divided code snippets with corresponding data subsets, accumulating results that collectively represent the full dataset analysis, thereby maintaining reliability while improving efficiency

Inventive Principle:

Principle #20Continuity of useful action

Solution Approach 2:

The patent applies partial action by processing divided portions of the training data through dynamic slicing rather than attempting to process the entire dataset in one operation. Each code snippet is processed with its corresponding data subset, and the results are aggregated to achieve comprehensive analysis, balancing speed and reliability

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20240232703A9Computer-readable recording medium storing output program, output method, and information processing apparatus

Publication Date: 2024.07.11 FUJITSU LTD

AI summary

A non-transitory computer-readable recording medium storing an output program causing a computer to execute a process, the process includes extracting an operation description related to a data operation for input data to a machine learning program, from the machine learning program, determining an extraction condition of target data to be operated in the data operation, based on the extracted operation description, extracting the target data which satisfies the determined extraction condition, from the input data, and outputting sampling data which includes the extracted target data.