Machine Learning Development Support System for Inference Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning engineers face challenges in efficiently applying machine learning programs and models developed by data scientists to application destination systems due to differences in data acquisition conditions, which affect inference accuracy, and lack of sufficient development records for modification.
Innovation Solution
A machine learning development support system that includes a data storage unit, data attribute extraction unit, experiment record information storage unit, and processing specification unit to extract and analyze data acquisition conditions and improve inference accuracy by referencing experiment records and modifying the machine learning program accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the machine learning engineer manually reads and analyzes the program code and experiment records to understand processing content and data acquisition conditions, then the engineer can grasp the development details, but the task burden increases significantly and efficiency decreases
Solution Approach 1:
The system automatically extracts and stores data acquisition conditions and processing content during the model development phase, before the engineer needs to apply the model. This preliminary extraction and storage of information eliminates the need for manual code reading later, reducing task burden while maintaining complete development records.
2Loss of information
If the system stores detailed development records automatically during model creation, then information availability improves, but system complexity increases
Solution Approach 1:
The system uses a unified data structure that can store multiple types of information (data acquisition conditions, processing content, model parameters) in a standardized format. This multi-functional storage approach maintains complete development records while avoiding the need for separate complex storage systems for each type of information.
3Reliability
If the machine learning engineer needs to confirm differences in data acquisition conditions between development environment and application destination system, then inference accuracy can be maintained, but the time and effort required increases
Solution Approach 1:
The system automatically compares the stored data acquisition conditions from the development phase with the current application environment, providing feedback on any differences. This automated feedback mechanism ensures inference accuracy is maintained by identifying condition mismatches while eliminating manual verification time and effort.
4Quantity of substance
If sufficient development records are not stored during model creation, then storage requirements are reduced, but the engineer cannot efficiently modify the model for different applications
Solution Approach 1:
The system extracts only the essential development record information (data acquisition conditions and processing content) that is necessary for model modification and application. This selective extraction maintains ease of operation by providing relevant information without storing excessive data, balancing storage requirements with modification efficiency.
Data Source
AI summary
A machine learning development support system includes: a data storage unit that stores training data; a data attribute extraction unit that refers to definition information whereby a conversion condition for converting the training data into an attribute thereof is defined and extracts data acquisition conditions from the training data; and an experiment record information storage unit that stores experiment record information., The machine learning product including a machine learning program and a trained model is divided into versions and recorded, and an accuracy of an inference obtained by each version is recorded in association with the version, the trained model, and the training data; and a processing specification unit that refers to the experiment record information, and specifies, for each of the data acquisition conditions, a machine learning product in which the accuracy of the inference is improved between old and new versions of the machine learning product.