1. Performance of Domain-Wall Encoding in a Digital Ising Machine
- Author
-
Kikuchi, Shuta, Takahashi, Kotaro, and Tanaka, Shu
- Subjects
Condensed Matter - Statistical Mechanics - Abstract
To tackle combinatorial optimization problems on an Ising machine, the objective function and constraints have to be mapped onto a quadratic unconstrained binary optimization (QUBO) model. Since the QUBO consists of binary variables when combinatorial optimization problems involve integer values, the integer values have to be encoded using binary variables. The encoding methods are called binary-integer encoding. Domain-wall encoding is one of the binary-integer encodings that has been recently proposed. Experiments conducted on a quantum annealing machine have demonstrated that the domain-wall encoding exhibits superior performance compared to the well-known one-hot encoding. In a digital Ising machine, the computation time required to reach the optimal or feasible solution was found to be shorter for domain-wall encoding than for one-hot encoding. However, the performance of domain-wall encoding on a digital Ising machine remains unclear from a practically important point of view. Therefore, we evaluated the performance of the one-hot encoding and domain-wall encoding in a digital Ising machine using the quadratic knapsack problem (QKP). To compare the performances, the penalty coefficient dependency and the sensitivity to the computation time were investigated. It is demonstrated that the domain-wall encoding indicated a higher feasible solution rate at the value of penalty coefficient, which was not commonly used in a number of previous studies. It is also shown that the domain-wall encoding gives higher values than the one-hot encoding in practical evaluation metrics for QKPs with large knapsack capacities. Furthermore, the domain-wall encoding is found to be more sensitive to computation time than the one-hot encoding., Comment: 10 pages, 7 figures
- Published
- 2024