Unconventional computing approaches have gained significant attention in the search for hardware-friendly solvers capable of finding the solution of combinatorial optimization problems COPs. Among others, oscillator-based, or oscillatory, Ising machines (OIMs), and simulated bifurcation-based Ising machines (SBMs) are two promising paradigms that are based on dynamical equations. This work presents a preliminary direct software comparison between an OIM and an SBM on instances from the G-set, a well-known benchmark library of maximum cut problems. The performance of the two paradigms is compared in terms of their average optimal approximation within a given number of timesteps using the same integration scheme. Our findings show that OIM performs better when optimizing its parameter for a specific problem or topology, while SBM has better performance in scenarios in which preprocessing is not possible, i.e., when considering large batches of instances with varied topologies and scales.
A comparison of Oscillatory Ising Machines and Simulated Bifurcation Machines for Solving Maximum Cut Problems
Grimaldi A.;Raimondo E.;Finocchio G.;
2024-01-01
Abstract
Unconventional computing approaches have gained significant attention in the search for hardware-friendly solvers capable of finding the solution of combinatorial optimization problems COPs. Among others, oscillator-based, or oscillatory, Ising machines (OIMs), and simulated bifurcation-based Ising machines (SBMs) are two promising paradigms that are based on dynamical equations. This work presents a preliminary direct software comparison between an OIM and an SBM on instances from the G-set, a well-known benchmark library of maximum cut problems. The performance of the two paradigms is compared in terms of their average optimal approximation within a given number of timesteps using the same integration scheme. Our findings show that OIM performs better when optimizing its parameter for a specific problem or topology, while SBM has better performance in scenarios in which preprocessing is not possible, i.e., when considering large batches of instances with varied topologies and scales.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.