Regression-Based Subspace Search for Design Variable Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optimizing design variables in complex systems, such as semiconductor apparatuses, is challenging due to the risk of achieving local optimal solutions when adjusting multiple influencing variables independently, and existing low-dimensional search methods rely heavily on designer selection of variable combinations, which can lead to inefficiencies and increased time costs.
Innovation Solution
An information processing apparatus that generates regression models to identify strongly interacting design variables, forms subgroups based on these interactions, and performs subspace searches to optimize design variables efficiently, reducing the likelihood of local optimal solutions by systematically combining and adjusting multiple variables simultaneously.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If low-dimensional search is used to reduce the number of variables considered at the same time, then optimization efficiency is improved, but the risk of leading to local optimal solutions increases
Solution Approach 1:
The patent segments the design variables into multiple subgroups, where each subgroup contains variables that strongly interact with each other. This segmentation allows the optimization process to handle variables in manageable chunks while preserving the critical interactions within each subgroup, thus maintaining optimization efficiency while reducing the risk of local optimal solutions.
Solution Approach 2:
The patent introduces an interaction strength calculation mechanism as an intermediary to objectively determine the grouping of variables. This intermediary uses regression analysis to quantify the interaction strength between variables, enabling automated and objective subgroup formation rather than relying on designer intuition, thereby improving both efficiency and reliability.
2Ease of operation
If design variables are selected by designer intuition for low-dimensional search, then the process is simple to implement, but inappropriate combinations may be selected leading to local optimal solutions
Solution Approach 1:
The system performs self-service by automatically calculating interaction strengths between variables using regression analysis and autonomously generating subgroups based on these calculations. This eliminates the need for designer intuition in variable selection, ensuring appropriate combinations are formed objectively while maintaining ease of operation through automated processes.
Solution Approach 2:
The patent replaces the mechanical process of designer intuition and manual variable selection with an automated computational mechanism. The regression-based interaction strength calculation and automatic subgroup generation substitute for human judgment, providing more reliable and consistent variable combinations without sacrificing operational simplicity.
3Ease of operation
If random selection of design variables is used for subgroups, then implementation is straightforward, but the number of searches increases leading to higher time costs
Solution Approach 1:
The patent replaces random selection with a deterministic mechanism based on interaction strength calculations. By using regression analysis to objectively identify which variables strongly interact, the system generates meaningful subgroups that require fewer searches to achieve effective optimization, thereby reducing time loss while maintaining ease of implementation through automation.
Solution Approach 2:
The patent changes the selection criterion from random to interaction-strength-based. By calculating and utilizing the interaction strength parameter between variables, the system transforms the variable selection process into one that is both efficient and time-effective, avoiding the need for numerous random searches while preserving implementation simplicity.
4Reliability
If the number of design variables to be adjusted is increased to avoid local optimal solutions, then optimization accuracy is improved, but the time cost increases
Solution Approach 1:
The patent segments all design variables into multiple subgroups based on their interaction strengths, allowing the optimization process to consider variables in focused groups rather than all at once. This segmentation enables the system to achieve high optimization accuracy by capturing strong interactions within subgroups while controlling the number of variables active in each search step, thereby reducing overall time cost.
Solution Approach 2:
The patent changes the approach from adjusting all variables simultaneously to adjusting variables in interaction-strength-based subgroups. This parameter change in the optimization strategy allows the system to achieve high accuracy by focusing computational resources on the most strongly interacting variable combinations, significantly reducing the time cost compared to exhaustive search methods.
Data Source
AI summary
An information processing apparatus according to one embodiment, comprising: a regression model generator configured to, by combining two or more of a plurality of variables, generate a plurality of terms that include combinations of two or more of the plurality of variables, respectively, and generate a regression model that regresses a property variable or an objective variable indicating an output of an objective function that includes the property variable, by the plurality of terms; a subgroup generator configured to generate, based on coefficients of the plurality of terms included in the regression model, subgroups that are the combinations of variables included the terms, respectively; and a subspace search processor configured to perform search for each of subspaces spanned by the subgroups based on an optimization criterion for the objective function, and generate pieces of first design value data that include values of the plurality of variables for the subspaces.