Combinatorial Black-box Optimization for Vehicle Design Problem

Abstract: Black-box optimization minimizes an objective function without derivatives or explicit forms. Such an optimization method with continuous variables has been successful in the fields of machine learning and material science. For discrete variables, the Bayesian optimization of combinatorial structure (BOCS) is a powerful tool for solving black-box optimization problems. A surrogate model used in BOCS is the quadratic unconstrained binary optimization (QUBO) form. Because of the approximation of the objective function to the QUBO form in BOCS, BOCS can expand the possibilities of using D-Wave quantum annealers, which can generate near-optimal solutions of QUBO problems by utilizing quantum fluctuation. We demonstrate the use of BOCS and its variant for a vehicle design problem, which cannot be described in the QUBO form. As a result, BOCS and its variant slightly outperform the random search, which randomly calculates the objective function.