With advancements in automated experimental setups, material optimisation and discovery can scale to higher throughput with larger evaluation budgets. Two state-of-the-art algorithms with conceptually different multi-objective optimisation strategies (Bayesian and Evolutionary) are compared on synthetic and real-world datasets. Our results show that the Bayesian optimisation strategy, q-Noisy Expected Hypervolume Improvement (qNEHVI) is superior in finding solutions at the Pareto Front rapidly, and when considering hypervolume improvement as a performance indicator. On the other hand, the Evolutionary optimisation strategy, Unified Non-dominated Sorting Genetic Algorithm III (U-NSGA-III), can exploit the Pareto Front and propose a larger pool of optimal solutions, given sufficient evaluation budget, and thus may be a better choice for materials discovery problems where knowing the complete Pareto Front provides greater scientific value to understanding materials space. We discuss the limitations of using hypervolume as a performance indicator for optimisation strategies, alongside hypervolume-based strategies such as qNEHVI, which do not adequately explain the number of solutions at or near the Pareto Front. We also performed a comparison of both optimisation strategies at different batch sizes to consider throughput capabilities.

With advancements in automation and high-throughput techniques, complex materials discovery with multiple conflicting objectives can now be tackled in experimental labs. Given that physical experimentation is greatly limited by evaluation budget, maximizing efficiency of optimization becomes crucial. We discuss the limitations of using hypervolume as a performance indicator for desired optimality across the entire multi-objective optimization run and propose new metrics specific to experimentation: ability to perform well for complex high-dimensional problems, minimizing wastage of evaluations, consistency/robustness of optimization, and ability to scale well to high throughputs. With these metrics, we perform a comparison of two conceptually different and state-of-the-art algorithms (Bayesian and Evolutionary) on synthetic and real-world datasets. We discuss the merits of both approaches with respect to exploration and exploitation, where fully resolving the Pareto Front could be the main aim for greater scientific value in understanding materials space, and thus provide a perspective for materials scientists to implement optimization in their platforms.

Supporting InformationYee-Fun Lim,* Chee Koon Ng, US Vaitesswar, Kedar Hippalgaonkar*