Output Space Entropy Search Framework for Multi-Objective Bayesian Optimization

We consider the problem of black-box multi-objective optimization (MOO) using expensive function evaluations (also referred to as experiments), where goal is approximate true Pareto set solutions by minimizing total resource cost experiments. For example, in hardware design optimization, we need find designs that trade-off performance, energy, and area overhead computational simulations. The key challenge select sequence experiments uncover high-quality minimal resources. In this paper, propose a general framework for solving MOO problems based on principle output space entropy (OSE) search: experiment maximizes information gained per unit about front. appropriately instantiate OSE search derive efficient algorithms following four settings: 1) most basic em single-fidelity setting, are accurate; 2) Handling constraints} which cannot be evaluated without performing experiments; 3) discrete multi-fidelity can vary amount resources consumed their evaluation accuracy; 4) continuous-fidelity continuous approximations result huge Experiments diverse synthetic real-world benchmarks show our improve over state-of-the-art methods terms both computational-efficiency accuracy solutions.