Greed is Good: Exploration and Exploitation Trade-offs in Bayesian Optimisation
This work addresses the exploration-exploitation trade-off in Bayesian optimization, offering improved strategies for practitioners in fields like computational fluid dynamics and robotics, though it is incremental as it builds on existing acquisition functions.
The paper investigated acquisition functions for Bayesian optimization to find global optima of continuous functions, showing that ε-greedy algorithms perform at least as well as conventional methods like Expected Improvement and Upper Confidence Bound, especially with limited budgets and in higher dimensions, as validated on benchmark problems and real-world applications.
The performance of acquisition functions for Bayesian optimisation to locate the global optimum of continuous functions is investigated in terms of the Pareto front between exploration and exploitation. We show that Expected Improvement (EI) and the Upper Confidence Bound (UCB) always select solutions to be expensively evaluated on the Pareto front, but Probability of Improvement is not guaranteed to do so and Weighted Expected Improvement does so only for a restricted range of weights. We introduce two novel $ε$-greedy acquisition functions. Extensive empirical evaluation of these together with random search, purely exploratory, and purely exploitative search on 10 benchmark problems in 1 to 10 dimensions shows that $ε$-greedy algorithms are generally at least as effective as conventional acquisition functions (e.g., EI and UCB), particularly with a limited budget. In higher dimensions $ε$-greedy approaches are shown to have improved performance over conventional approaches. These results are borne out on a real world computational fluid dynamics optimisation problem and a robotics active learning problem. Our analysis and experiments suggest that the most effective strategy, particularly in higher dimensions, is to be mostly greedy, occasionally selecting a random exploratory solution.