Optimization Solver Penalty Tuning for Constraint Satisfaction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optimization solvers, such as Ising machines, struggle with setting constraint conditions and handling high-order objective functions, leading to a low probability of outputting high-quality solutions that satisfy constraints or match the original problem's objective function.
Innovation Solution
An information processing device that converts optimization problems into approximate problems using penalty terms and sampling-based learning to generate unconstrained functions, which are solved by a simulated bifurcation machine, ensuring high-quality solutions are efficiently calculated while satisfying constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a strong coupling coefficient is used in the penalty term, then the probability of satisfying constraint conditions increases, but the quality of the solution decreases
Solution Approach 1:
The patent dynamically adjusts the coupling coefficient of the penalty term during the optimization process. Initially, a stronger penalty is applied to enforce constraint satisfaction, and then the penalty strength is gradually reduced to allow for higher solution quality. This time-varying approach resolves the contradiction by making the constraint enforcement flexible rather than static.
Solution Approach 2:
The patent first transforms the original problem into an approximate problem with penalty terms that enforce constraint satisfaction before solving. This preliminary transformation ensures constraints are built into the optimization landscape from the start, allowing the solver to find feasible solutions that can then be refined for quality.
2Manufacturing precision
If a weak coupling coefficient is used in the penalty term, then the solution quality increases, but the probability of satisfying constraint conditions decreases
Solution Approach 1:
The patent employs dynamic adjustment of the penalty coupling coefficient, transitioning from strong initial penalties to weaker final penalties. This allows the system to first ensure constraint satisfaction and then optimize for solution quality, resolving the trade-off between reliability and precision.
Solution Approach 2:
The optimization process uses periodic or staged adjustment of the penalty term strength, alternating between phases of constraint enforcement and solution refinement. This periodic modulation of penalty strength allows the system to achieve both constraint satisfaction and high solution quality at different stages.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
According to an arrangement, an information processing device (20) is configured to output a solution in a target optimization problem for minimizing or maximizing a target objective function, and includes a processing unit (401) configured to: provide a solver device configured to execute a solution-finding process while iteratively updating intermediate states of a solution, with an approximate problem for minimizing or maximizing an approximate function based on the target objective function ; acquire a plurality of intermediate solutions based on the intermediate states and a final solution calculated after the plurality of intermediate solutions; calculate, for each of the plurality of intermediate solutions and the final solution, an evaluation value as a solution to the target optimization problem; and output, as the solution to the target optimization problem, any selected solution selected from among the plurality of intermediate solutions and the final solution based on the evaluation value.