Hyperparameter Group Management via Base and Difference Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of managing hyperparameters in machine learning model training is high due to the large number of parameters and their dependency relationships, especially when using external files for hyperparameter definition, which becomes cumbersome and difficult to manage across multiple trials.
Innovation Solution
A setting program and information processing apparatus that generate and manage hyperparameter groups by combining base and difference parameters, allowing users to select and combine these parameters through a graphical user interface or command line interface, reducing the need for manual creation of external files and simplifying the management of hyperparameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If hyperparameters are defined in external files for each trial, then trial reproducibility is improved, but device complexity increases due to managing numerous parameter files and their dependencies
Solution Approach 1:
The patent segments hyperparameter management into base parameters (defined once) and difference parameters (applied as variations). This divides the complex task of managing complete hyperparameter sets into manageable components, where a base hyperparameter group can be segmented and combined with multiple difference parameter groups to generate numerous trial configurations without creating separate files for each trial.
Solution Approach 2:
The patent merges base parameters with difference parameters to generate complete hyperparameter groups for multiple trials. Instead of maintaining separate external files for each trial, the system combines a single base parameter definition with multiple difference parameter definitions, automatically generating the necessary hyperparameter configurations through programmatic combination.
2Adaptability or versatility
If multiple external files are created for different trials, then trial-specific parameter customization is improved, but loss of time increases due to manual file creation and management
Solution Approach 1:
The patent performs preliminary action by defining base parameters once and pre-defining multiple difference parameter groups. This preliminary setup enables automatic generation of trial-specific hyperparameter configurations without manual file creation for each trial, significantly reducing the time required for parameter management while preserving customization capabilities.
Solution Approach 2:
The patent uses copying by generating trial-specific hyperparameter configurations through programmatic combination of base and difference parameters. Instead of manually creating and copying parameter files for each trial, the system automatically generates configurations by combining parameter definitions, reducing manual effort while maintaining trial-specific customization.
3Manufacturing precision
If all hyperparameters are managed individually, then parameter precision control is improved, but device complexity increases due to the large number of parameters and their relationships
Solution Approach 1:
The patent segments the large number of hyperparameters into base parameters and difference parameters. This segmentation maintains precise control over individual parameter settings while reducing the perceived complexity by organizing parameters into logical groups with specific functions (base vs. difference parameters).
Data Source
AI summary
A non-transitory computer-readable recording medium stores a setting program for causing a computer to execute a process including: accepting designation of a first setting value for an entirety of a plurality of hyperparameters included in a machine learning model and designation of one or more second setting values that respectively indicate differences with respect to the first setting value; and causing a plurality of hyperparameter groups generated based on the first setting value that has been designated and the second setting values that have been designated to be set in the machine learning model.