Ising Machine Parallel Subproblem Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ising machines face challenges in storing and processing large numbers of weight coefficients when dealing with large-scale discrete optimization problems, leading to inefficiencies in calculation and potential memory overflow issues.
Innovation Solution
The optimization apparatus employs a plurality of Ising machines that divide a problem into subproblems and process them in parallel, with each machine storing values of state variables for multiple groups and calculating second solution candidates based on first solution candidates from other machines, ensuring efficient use of arithmetic units and memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the problem is divided into subproblems and processed in parallel by multiple Ising machines, then the calculation efficiency is improved, but the device complexity increases
Solution Approach 1:
The patent divides a large-scale optimization problem into multiple subproblems, with each subproblem assigned to a separate Ising machine. This segmentation allows parallel processing of subproblems, improving overall calculation efficiency while managing the complexity of handling large numbers of state variables and weight coefficients across multiple machines.
2Quantity of substance
If weight coefficients are stored in a large-capacity storage apparatus, then the memory capacity is sufficient, but the processing time increases
Solution Approach 1:
The patent segments the weight coefficients into multiple groups, with each group stored in the internal memory of a specific Ising machine. This allows each machine to access its assigned weight coefficients quickly without requiring transposition or transfer from external storage, significantly reducing processing time while maintaining sufficient memory capacity through distributed storage across multiple machines.
3Speed
If the same Ising model is assigned to each of multiple Ising chips for parallel processing, then the processing speed is improved, but the adaptability decreases
Solution Approach 1:
The patent assigns different subproblems with different Ising models to each Ising machine based on the specific characteristics and requirements of each subproblem. This allows each machine to be optimized for its specific task while maintaining parallel processing capabilities, thereby preserving adaptability across the system even as processing speed increases through parallelization.
Data Source
AI summary
An optimization apparatus includes: Ising machines that each search for a solution of a subproblem among subproblems obtained by dividing a problem represented by an Ising model, each of the Ising machines including: a memory configured to store values of a state variable set corresponding to a subproblem to be calculated for a plurality of groups; a processor configured to: perform processing of calculating a second solution candidate for each of the plurality of groups based on first solution candidates of subproblems calculated by other Ising machines by changing a group to be processed for every unit period such that the second solution candidate is calculated for the state variable set of a group different from groups processed by the other Ising machines in a same unit period; and a communication circuit configured to transmit the second solution candidate to the other Ising machines every time the unit period ends.