| Summary: | With advancements in automation and high-throughput techniques, we can tackle more complex multi-objective materials discovery problems requiring a higher evaluation budget. Given that experimentation is greatly limited by evaluation budget, maximizing sample efficiency of optimization becomes crucial. We discuss the limitations of using hypervolume as a performance indicator and propose new metrics relevant to materials experimentation: such as the 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 an empirical study 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 provides more knowledge of the best material.
|
|---|