Parameter Estimation With Dimension-Reduced Bayesian Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Bayesian optimization techniques face challenges with high computational costs, especially in high-dimensional parameter spaces, due to shared search information among workers in parallel processing, dimension reduction without parallelization, and time-consuming space division methods.
Innovation Solution
A parameter estimation system that divides the searching range by reducing dimensions and allocates these ranges to multiple workers for asynchronous parallel processing, allowing for faster determination of optimum input parameters without additional computation costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If Bayesian optimization is performed in parallel with multiple workers, then the search speed is improved, but the search information is shared among workers causing delays
Solution Approach 1:
The parameter space is divided into multiple subspaces, and each worker is assigned to search a specific subspace independently. This segmentation eliminates the need for workers to share and wait for search information from other workers, as each worker operates autonomously on its own assigned subspace while contributing to the overall optimization goal.
2Productivity
If the parameter space is divided to optimize each section, then the search efficiency is improved, but the time required for division becomes relatively large
Solution Approach 1:
The parameter space is divided into subspaces in advance, before the actual optimization search begins. This preliminary segmentation allows workers to immediately start searching their assigned subspaces without incurring division delays during the optimization process, thus improving overall search efficiency while minimizing the impact of division time.
3Quantity of substance
If dimension reduction is performed, then the number of observation points is reduced, but the dimension number must be equal to or greater than the effective dimension number
Solution Approach 1:
Instead of reducing the overall dimension of the parameter space, the method segments the high-dimensional space into multiple lower-dimensional subspaces. Each worker performs optimization in a reduced-dimensional subspace, achieving fewer observation points per worker while maintaining sufficient dimensionality (equal to or greater than the effective dimension number) within each subspace to preserve optimization accuracy.
Data Source
AI summary
The present invention relates to a parameter estimation system, a parameter estimation method, and a program, and more particularly to a parameter estimation system, a parameter estimation method, and a program that efficiently estimate parameters of machine learning and simulation, etc. An objective of the present invention is to provide a parameter estimation system and a parameter estimation method that may rapidly determine the optimum input parameter.