Parameter Setting Robustness via Evaluation Variation Estimation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing techniques for adjusting parameter settings in simulations or tests often result in deviations from the optimal set value due to various factors, leading to substantial effects on the system or product.

Innovation Solution

An information processing apparatus comprising an estimator and a recommender, which estimates the relationship between set values and evaluation value variations, and recommends a set value that minimizes the effect of deviations, thereby providing a robust and stable optimum value.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If Bayesian optimization is used to adjust parameter set values, then the number of trials required for adjustment is reduced, but the setting may deviate from the optimal value due to various factors

Engineering Contradiction:

Improvenumber of trials for parameter adjustmentVSAvoidaccuracy of parameter setting

Core Design Contradiction:

Loss of timeVSReliability

Solution Approach 1:

The patent changes the optimization parameter from minimizing evaluation value to minimizing evaluation value variation. This parameter transformation allows the system to identify set values that are robust against deviations, thereby improving reliability while maintaining efficiency in the number of trials required

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent creates a virtual model (surrogate model) that copies the relationship between set values and evaluation values. This virtual model allows the system to predict evaluation value variations without conducting numerous actual trials, thus reducing time loss while improving the reliability of parameter settings through virtual experimentation

Inventive Principle:

Principle #26Copying

2Productivity

If a set value regarded as optimum is used, then the system operates efficiently, but the setting may deviate from the optimal value leading to substantial effects

Engineering Contradiction:

Improvesystem operation efficiencyVSAvoidstability of parameter setting

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies beforehand cushioning by pre-identifying set values that have small evaluation value variations before actual system operation. By selecting set values in advance that are robust against deviations, the system maintains both high productivity and reliability during operation, cushioning against potential deviations from optimal values

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

Data Source

PatentUS12223236B2Information processing apparatus, information processing method, and non-transitory computer readable medium

Publication Date: 2025.02.11 KK TOSHIBA

AI summary

An information processing apparatus according to an embodiment of the present invention includes an estimator and a recommender. The estimator is configured to, based on a data set including a set value set for a parameter and an evaluation value or an evaluation value variation where the set value is set for the parameter, estimate a relationship between the set value and the evaluation value variation. The evaluation value variation indicates a variation of respective evaluation values where a plurality of values included within a neighborhood range that is based on the set value are set for the parameter. The recommender is configured to, based on the estimated relationship, determine a recommended set value.