Efficient Ground-state Searches by Scheduling Sparsity of Interactions of Physical Spin Dynamics for Ising Spin Computing

Abstract

Ising spin computing has received gaining attention as an efficient computing technology for solving combinatorial optimization problems. We have introduced extraction-type majority voting logic (E-MVL) that purposely disconnects the interactions between spins and controls the sparsity to find the ground state. In this study, we examine how to control the sparsity, which is a key factor in performance. As a result, the residual energy of E-MVL is reduced by 33.2% compared with that of highly optimized simulated annealing (SA) at solving the Sherrington-Kirkpatrick model with 400 spins. Further, we show that E-MVL exhibits acceleration by increasing the sparsity. These results indicate that E-MVL provides faster and more accurate optimizations than SA by setting an appropriate sparsity.