Continuous black-box optimization with an Ising machine and random subspace coding

A black-box optimization algorithm such as Bayesian finds the extremum of an unknown function by alternating inference underlying and acquisition function. In a high-dimensional space, algorithms perform poorly due to difficulty optimization. Herein, we apply Ising machines overcome in continuous As machine specializes binary problems, vector has be encoded binary, solution translated back. Our method following three parts: (1) random subspace coding based on axis-parallel hyperrectangles from vector, (2) quadratic unconstrained (QUBO) defined nonnegative-weighted linear regression model, which is solved machines, (3) penalization scheme ensure that can It shown with benchmark tests its performance using D-Wave Advantage quantum annealer simulated annealing competitive state-of-the-art Gaussian process problems. may open up possibility other QUBO solvers, including approximate gated-quantum computers, expand range application continuous-valued