Ising Model Optimization Apparatus with Segmented State Variable Groups
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optimization apparatuses face challenges in efficiently calculating problems with 1-hot constraints while maintaining hardware complexity at a manageable level, often resulting in a larger search space and complex energy calculation hardware.
Innovation Solution
The optimization method involves an apparatus with multiple storage units and a searching unit that performs update processing by accessing weight coefficient groups to determine energy changes and update state variables, ensuring constraints are met, thereby reducing hardware complexity and narrowing the search space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the apparatus performs exhaustive energy calculations for all state variable combinations to ensure constraint satisfaction, then calculation accuracy is improved, but hardware complexity and calculation time increase significantly
Solution Approach 1:
The patent segments the state variables into multiple groups (first state variable group, second state variable group, etc.) and performs energy calculations for each group separately. The searching unit calculates energy changes by updating one group at a time while keeping other groups fixed, thereby reducing the computational burden on hardware compared to evaluating all state variables simultaneously.
Solution Approach 2:
The apparatus performs preliminary energy calculations to determine whether state variable updates should be performed before actually executing the updates. The searching unit calculates energy changes in advance for potential updates, and only performs updates when the energy change indicates improvement, thereby avoiding unnecessary complex calculations later.
2Productivity
If the apparatus updates all state variables simultaneously to explore the search space efficiently, then search speed is improved, but the difficulty of maintaining constraint satisfaction increases
Solution Approach 1:
The patent divides state variables into multiple groups and updates them sequentially rather than simultaneously. The searching unit updates the first state variable group, then the second state variable group, and so on, while maintaining constraint satisfaction within each group. This segmentation approach enables parallel processing of different groups while simplifying constraint management within each group.
Solution Approach 2:
The apparatus performs partial updates by updating only one state variable group at a time rather than all state variables simultaneously. This partial action approach maintains constraint satisfaction more easily while still making progress through the search space, and can be repeated across multiple iterations to achieve comprehensive search.
3Device complexity
If the apparatus reduces the number of energy calculations to simplify hardware, then hardware complexity is reduced, but the quality of optimization results deteriorates
Solution Approach 1:
The searching unit performs preliminary energy calculations to determine whether state variable updates should be performed before actually executing the updates. By calculating energy changes in advance and using these results to guide update decisions, the apparatus ensures that optimization quality is maintained while avoiding unnecessary calculations that would increase hardware complexity.
Solution Approach 2:
The apparatus uses the energy calculation results to automatically determine whether updates should be performed, eliminating the need for complex external control mechanisms. The searching unit self-regulates the update process based on energy change criteria, thereby maintaining optimization quality with simpler hardware compared to systems requiring external control for each calculation decision.
Data Source
AI summary
A method includes: accessing first storage configured to store a first weight coefficient group which is at least some of a plurality of weight coefficients indicating a magnitude of interaction between a plurality of state variables in an evaluation function representing energy of an Ising model; accessing a plurality of second storages each of the plurality of second storage being configured to store a second weight coefficient group related to a state variable having a value of 1 in any of a plurality of state variable groups respectively including the plurality of state variables among the plurality of weight coefficients; outputting, for each of the plurality of state variable groups, a search result obtained by performing searching processing configured to perform processing of searching for an optimum solution by repeatedly performing a first update process with a first constraint or a second update process with a second constraint.