Ising Optimization Device Using Four-State 2-Way 1-Hot Transitions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional optimization devices face challenges in efficiently solving optimization problems with 2-Way 1-hot constraints due to the occurrence of state transitions that do not satisfy these constraints, leading to increased search spaces and longer calculation times.
Innovation Solution
The optimization device models the problem in an Ising model and allows simultaneous changes in four state variables to satisfy the 2-Way 1-hot constraint, using a propagation unit to update local fields and weighting coefficients, and a transition selection unit to probabilistically accept changes that reduce the search space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If only one state variable changes at a time (Hamming distance = 1), then the state transition is simple to control, but the search space is unnecessarily large and calculation time increases
Solution Approach 1:
The patent segments the state transition process into two distinct phases: (1) candidate generation phase where multiple potential state changes are identified simultaneously, and (2) selection phase where the optimal transition is chosen based on energy change and constraint satisfaction. This segmentation allows the system to evaluate multiple transitions in parallel while maintaining controlled complexity through structured candidate generation and selection criteria.
2Adaptability or versatility
If state transitions are allowed freely, then the search space coverage is comprehensive, but transitions violating the 2-Way 1-hot constraint occur frequently
Solution Approach 1:
The patent applies preliminary action by pre-calculating and identifying candidate state transitions that satisfy the 2-Way 1-hot constraint before executing the actual state change. The constraint checking is performed in advance during the candidate generation phase, ensuring that only valid transitions are considered. This preliminary validation prevents constraint violations rather than correcting them after occurrence.
3Measurement precision
If the search space includes all possible states, then the optimization solution is thorough, but the number of states not satisfying the 2-Way 1-hot constraint increases the calculation time
Solution Approach 1:
The patent applies local quality by making the search process adaptive to local constraint conditions. Instead of uniformly exploring all possible state transitions, the system identifies and focuses computational resources on transitions that satisfy the 2-Way 1-hot constraint in each local region of the state space. The candidate generation mechanism dynamically adjusts the search focus based on local constraint satisfaction, thereby improving calculation efficiency without sacrificing solution accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method includes: calculating a change amount of energy of an Ising model, the change amount being caused when values of four states are changed, for each of a plurality of first states having a value of 0 before change, so as to satisfy a constraint that a sum of values of states included in each row is 1, and a sum of values of states included in each column is 1, when N (an integer of 2 or larger) squared state variables included in an evaluation function representing the energy of the Ising model are arrayed in N rows and N columns; selecting the four states allowed to change by using the change amount calculated for each of the plurality of first states, and outputting identification information for identifying the four states; and updating values of the four states allowed to change by using the identification information.