Dynamic Ising Model Convergence Judgment for Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for solving large-scale discrete optimization problems, such as those using Ising machines, face challenges in determining the optimal number of iterations for convergence judgment, leading to either unnecessary prolonged searches or premature termination, which can result in overlooking the best parameter values.
Innovation Solution
An information processing method that dynamically determines the number of iterations for convergence judgment based on the difference between evaluation values, the number of bits, and the types and numbers of constraints in the evaluation function, using formulas to adjust the iteration count and ensure accurate parameter search termination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a fixed number of iterations is used for convergence judgment, then the parameter search process is simple to implement, but it leads to either unnecessary prolonged searches or premature termination
Solution Approach 1:
The patent applies dynamics by making the iteration count dynamic rather than fixed. The convergence judgment iteration count is determined based on problem characteristics (number of bits, constraint types and numbers) and evaluation function properties, allowing the system to adapt the number of iterations to each specific problem instance, thereby avoiding both unnecessary prolongation and premature termination of searches
Solution Approach 2:
The patent changes parameters by using multiple formulas to determine different iteration counts based on various problem characteristics. Different formulas are applied depending on the specific problem type, transforming the single fixed iteration parameter into multiple adjustable parameters that optimize the search process for different scenarios
2Ease of operation
If a fixed number of iterations is used for convergence judgment, then the implementation is straightforward, but it may overlook the best parameter values
Solution Approach 1:
The patent makes the convergence judgment adaptive by determining iteration counts dynamically based on problem characteristics. This ensures that the search process continues long enough to find the best parameter values while avoiding premature termination, thereby improving the accuracy of parameter search without complicating the implementation
Solution Approach 2:
The patent implements feedback mechanisms by using evaluation functions to assess parameter quality and determining convergence based on this feedback. The system monitors evaluation results during the search process and uses this information to judge convergence, ensuring that the best parameter values are not overlooked while maintaining a systematic approach
3Measurement precision
If different formulas are used to determine iteration counts based on problem characteristics, then the parameter search accuracy is improved, but the device complexity increases
Solution Approach 1:
The patent applies local quality by using different formulas for different problem characteristics. Each formula is optimized for specific problem types (based on number of bits, constraint types, and constraint numbers), allowing the system to apply the most appropriate calculation method locally rather than using a single complex formula universally
Solution Approach 2:
The patent segments the convergence judgment process into multiple formulas that handle different problem characteristics separately. By dividing the determination process into distinct calculation paths based on problem features, the system improves accuracy for each segment while keeping the overall structure manageable through modular formula selection
Data Source
AI summary
An information processing method for a computer to execute a process includes searching for a first parameter for searching for a solution by using an evaluation function included in an Ising model; acquiring an evaluation value of the first parameter based on an energy value of the evaluation function and a time period to reach the energy value; determining a number of iterations for determining that the searching the first parameter converges based on at least one selected from a best evaluation value after the searching the first parameter for a certain number of iterations, a number of bits included in the evaluation function, a type of a constraint, and a number of the constraint; and when the best evaluation value is not updated during the searching for the first parameter for the number of iterations, determining a second parameter which leads to the best evaluation value.